diff --git a/benchmarks/4c32ab6/benchmark.csv b/benchmarks/4c32ab6/benchmark.csv
new file mode 100644
index 00000000..87327c70
--- /dev/null
+++ b/benchmarks/4c32ab6/benchmark.csv
@@ -0,0 +1,2485 @@
+kernel_name,kernel_provider,kernel_operation_mode,metric_name,metric_unit,x_name,x_label,x_value,y_value_50,y_value_20,y_value_80,extra_benchmark_config_str,gpu_name,timestamp,liger_version
+fused_linear_cpo_loss,liger,forward,speed,ms,B,B,2,11.842319965362549,11.697619247436524,11.906118202209473,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:12:20,0.6.3
+fused_linear_cpo_loss,liger,forward,speed,ms,B,B,4,23.48263931274414,23.433970642089847,23.496837615966797,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:12:20,0.6.3
+fused_linear_cpo_loss,liger,forward,speed,ms,B,B,8,47.82439994812012,47.699090576171876,47.94970932006836,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:12:20,0.6.3
+fused_linear_cpo_loss,liger,forward,speed,ms,B,B,16,95.93929290771484,95.93929290771484,95.93929290771484,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:12:20,0.6.3
+fused_linear_cpo_loss,huggingface,forward,speed,ms,B,B,2,6.02675199508667,6.023039817810059,6.043039798736572,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:12:46,0.6.3
+fused_linear_cpo_loss,huggingface,forward,speed,ms,B,B,4,11.791855812072754,11.78116455078125,11.837772750854493,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:12:46,0.6.3
+fused_linear_cpo_loss,huggingface,forward,speed,ms,B,B,8,23.140064239501953,23.137586975097655,23.143104553222656,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:12:46,0.6.3
+fused_linear_cpo_loss,huggingface,forward,speed,ms,B,B,16,46.1693115234375,46.1632438659668,46.17537918090821,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:12:46,0.6.3
+fused_linear_cpo_loss,liger,backward,speed,ms,B,B,2,1.0881279706954956,1.0849535942077637,1.090937614440918,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:13:10,0.6.3
+fused_linear_cpo_loss,liger,backward,speed,ms,B,B,4,1.1098400354385376,1.1054464101791381,1.1248575925827027,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:13:10,0.6.3
+fused_linear_cpo_loss,liger,backward,speed,ms,B,B,8,1.1220000386238098,1.1194623947143554,1.1322495698928834,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:13:10,0.6.3
+fused_linear_cpo_loss,liger,backward,speed,ms,B,B,16,1.1702399849891663,1.1676735877990723,1.1807680130004883,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:13:10,0.6.3
+fused_linear_cpo_loss,huggingface,backward,speed,ms,B,B,2,10.969696044921875,10.967545509338379,10.98120937347412,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:13:35,0.6.3
+fused_linear_cpo_loss,huggingface,backward,speed,ms,B,B,4,21.03985595703125,21.005049896240234,21.054309844970703,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:13:35,0.6.3
+fused_linear_cpo_loss,huggingface,backward,speed,ms,B,B,8,40.68120002746582,40.679269409179696,40.68313064575195,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:13:35,0.6.3
+fused_linear_cpo_loss,huggingface,backward,speed,ms,B,B,16,82.4188461303711,82.4188461303711,82.4188461303711,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:13:35,0.6.3
+fused_linear_cpo_loss,liger,full,speed,ms,B,B,2,13.553600311279297,13.54753952026367,13.633689880371094,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:14:00,0.6.3
+fused_linear_cpo_loss,liger,full,speed,ms,B,B,4,25.182079315185547,25.1183162689209,25.215295791625977,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:14:00,0.6.3
+fused_linear_cpo_loss,liger,full,speed,ms,B,B,8,50.37926483154297,50.19210357666016,50.566426086425786,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:14:00,0.6.3
+fused_linear_cpo_loss,liger,full,speed,ms,B,B,16,95.22911834716797,95.22911834716797,95.22911834716797,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:14:00,0.6.3
+fused_linear_cpo_loss,huggingface,full,speed,ms,B,B,2,17.056480407714844,17.045018005371094,17.063801956176757,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:14:26,0.6.3
+fused_linear_cpo_loss,huggingface,full,speed,ms,B,B,4,33.06575965881348,33.00334014892578,33.12817916870117,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:14:26,0.6.3
+fused_linear_cpo_loss,huggingface,full,speed,ms,B,B,8,64.45439910888672,64.45439910888672,64.45439910888672,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:14:26,0.6.3
+fused_linear_cpo_loss,huggingface,full,speed,ms,B,B,16,133.2860107421875,133.2860107421875,133.2860107421875,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:14:26,0.6.3
+fused_linear_cpo_loss,liger,full,memory,MB,B,B,2,6142.03466796875,6142.03466796875,6142.03466796875,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:14:52,0.6.3
+fused_linear_cpo_loss,liger,full,memory,MB,B,B,4,6190.0517578125,6190.0517578125,6190.0517578125,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:14:52,0.6.3
+fused_linear_cpo_loss,liger,full,memory,MB,B,B,8,6286.0859375,6286.0859375,6286.0859375,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:14:52,0.6.3
+fused_linear_cpo_loss,liger,full,memory,MB,B,B,16,6478.154296875,6478.154296875,6478.154296875,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:14:52,0.6.3
+fused_linear_cpo_loss,huggingface,full,memory,MB,B,B,2,6610.017578125,6610.017578125,6610.017578125,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:19,0.6.3
+fused_linear_cpo_loss,huggingface,full,memory,MB,B,B,4,10148.033203125,10148.033203125,10148.033203125,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:19,0.6.3
+fused_linear_cpo_loss,huggingface,full,memory,MB,B,B,8,17226.064453125,17226.064453125,17226.064453125,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:19,0.6.3
+fused_linear_cpo_loss,huggingface,full,memory,MB,B,B,16,31382.126953125,31382.126953125,31382.126953125,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:19,0.6.3
+cross_entropy,liger,forward,speed,ms,V,vocab size,4096,0.4261920005083084,0.4185088038444519,0.43302398920059204,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:31,0.6.3
+cross_entropy,liger,forward,speed,ms,V,vocab size,8192,0.5435519814491272,0.5348607897758484,0.5492479801177979,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:31,0.6.3
+cross_entropy,liger,forward,speed,ms,V,vocab size,16384,1.033568024635315,1.0279232501983642,1.0428224325180053,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:31,0.6.3
+cross_entropy,liger,forward,speed,ms,V,vocab size,32768,1.6779199838638306,1.6701375722885132,1.69367036819458,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:31,0.6.3
+cross_entropy,liger,forward,speed,ms,V,vocab size,65536,4.022655963897705,4.014886569976807,4.030278205871582,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:31,0.6.3
+cross_entropy,liger,forward,speed,ms,V,vocab size,131072,8.610176086425781,8.606623649597168,8.613759994506836,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:31,0.6.3
+cross_entropy,huggingface,forward,speed,ms,V,vocab size,4096,0.24428799748420715,0.24352000653743744,0.24505600333213806,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:31,0.6.3
+cross_entropy,huggingface,forward,speed,ms,V,vocab size,8192,0.5492479801177979,0.547430419921875,0.55103999376297,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:31,0.6.3
+cross_entropy,huggingface,forward,speed,ms,V,vocab size,16384,0.7995520234107971,0.798412823677063,0.8005184173583985,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:31,0.6.3
+cross_entropy,huggingface,forward,speed,ms,V,vocab size,32768,2.630224108695984,2.625152111053467,2.6419200897216797,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:31,0.6.3
+cross_entropy,huggingface,forward,speed,ms,V,vocab size,65536,5.513376235961914,5.511532783508301,5.516793727874756,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:31,0.6.3
+cross_entropy,huggingface,forward,speed,ms,V,vocab size,131072,11.089327812194824,11.081190490722655,11.09368953704834,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:31,0.6.3
+cross_entropy,liger,backward,speed,ms,V,vocab size,4096,0.2417919933795929,0.23904000222682953,0.24422399699687958,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:32,0.6.3
+cross_entropy,liger,backward,speed,ms,V,vocab size,8192,0.4254719913005829,0.42047359347343444,0.4316799998283386,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:32,0.6.3
+cross_entropy,liger,backward,speed,ms,V,vocab size,16384,0.7848320007324219,0.7803327798843384,0.8061887860298157,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:32,0.6.3
+cross_entropy,liger,backward,speed,ms,V,vocab size,32768,1.486016035079956,1.4840576171875,1.4908352136611938,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:32,0.6.3
+cross_entropy,liger,backward,speed,ms,V,vocab size,65536,2.8927680253982544,2.8901888847351076,2.894585657119751,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:32,0.6.3
+cross_entropy,liger,backward,speed,ms,V,vocab size,131072,5.7163519859313965,5.711872100830078,5.718624114990234,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:32,0.6.3
+cross_entropy,huggingface,backward,speed,ms,V,vocab size,4096,0.374208003282547,0.37321600317955017,0.37571200728416443,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:33,0.6.3
+cross_entropy,huggingface,backward,speed,ms,V,vocab size,8192,0.7147200107574463,0.7136896014213563,0.7156864047050475,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:33,0.6.3
+cross_entropy,huggingface,backward,speed,ms,V,vocab size,16384,1.58433598279953,1.5831680297851562,1.5852479934692383,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:33,0.6.3
+cross_entropy,huggingface,backward,speed,ms,V,vocab size,32768,3.470560073852539,3.4695104122161866,3.4715840816497803,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:33,0.6.3
+cross_entropy,huggingface,backward,speed,ms,V,vocab size,65536,6.937615871429443,6.933830451965331,6.940038490295411,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:33,0.6.3
+cross_entropy,huggingface,backward,speed,ms,V,vocab size,131072,13.845104217529297,13.843084907531738,13.85010585784912,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:33,0.6.3
+cross_entropy,liger,full,speed,ms,V,vocab size,4096,0.7007039785385132,0.6870784044265748,0.7243263840675354,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:34,0.6.3
+cross_entropy,liger,full,speed,ms,V,vocab size,8192,1.0055360198020935,0.992793595790863,1.0269311904907226,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:34,0.6.3
+cross_entropy,liger,full,speed,ms,V,vocab size,16384,1.8423999547958374,1.832095980644226,1.8613439798355103,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:34,0.6.3
+cross_entropy,liger,full,speed,ms,V,vocab size,32768,3.1968319416046143,3.1815103054046627,3.212287998199463,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:34,0.6.3
+cross_entropy,liger,full,speed,ms,V,vocab size,65536,6.9747679233551025,6.963929462432861,6.9882945060729975,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:34,0.6.3
+cross_entropy,liger,full,speed,ms,V,vocab size,131072,14.383519649505615,14.374879837036133,14.3854398727417,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:34,0.6.3
+cross_entropy,huggingface,full,speed,ms,V,vocab size,4096,0.8766560256481171,0.8755071878433227,0.8777535915374755,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:35,0.6.3
+cross_entropy,huggingface,full,speed,ms,V,vocab size,8192,1.7821120023727417,1.780185651779175,1.7835583925247191,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:35,0.6.3
+cross_entropy,huggingface,full,speed,ms,V,vocab size,16384,3.418784022331238,3.41714563369751,3.4202432155609133,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:35,0.6.3
+cross_entropy,huggingface,full,speed,ms,V,vocab size,32768,8.176608085632324,8.166707229614257,8.188934135437012,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:35,0.6.3
+cross_entropy,huggingface,full,speed,ms,V,vocab size,65536,16.599647521972656,16.59591636657715,16.600473403930664,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:35,0.6.3
+cross_entropy,huggingface,full,speed,ms,V,vocab size,131072,33.25262451171875,33.24270706176758,33.262541961669925,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:35,0.6.3
+cross_entropy,liger,no-grad-forward,speed,ms,V,vocab size,4096,0.42204800248146057,0.4152384102344513,0.4305855929851532,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:36,0.6.3
+cross_entropy,liger,no-grad-forward,speed,ms,V,vocab size,8192,0.5541759729385376,0.5467455863952637,0.5640447854995728,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:36,0.6.3
+cross_entropy,liger,no-grad-forward,speed,ms,V,vocab size,16384,1.0416799783706665,1.0363263845443726,1.0462783813476562,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:36,0.6.3
+cross_entropy,liger,no-grad-forward,speed,ms,V,vocab size,32768,1.6792479753494263,1.6766079664230347,1.686527967453003,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:36,0.6.3
+cross_entropy,liger,no-grad-forward,speed,ms,V,vocab size,65536,4.036768198013306,4.02782096862793,4.041375827789307,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:36,0.6.3
+cross_entropy,liger,no-grad-forward,speed,ms,V,vocab size,131072,8.615072250366211,8.610879898071289,8.621472358703613,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:36,0.6.3
+cross_entropy,huggingface,no-grad-forward,speed,ms,V,vocab size,4096,0.24428799748420715,0.24355199933052063,0.24505600333213806,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:36,0.6.3
+cross_entropy,huggingface,no-grad-forward,speed,ms,V,vocab size,8192,0.5490239858627319,0.5473920106887817,0.5507455825805664,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:36,0.6.3
+cross_entropy,huggingface,no-grad-forward,speed,ms,V,vocab size,16384,0.7994880080223083,0.7985919713973999,0.8000959753990173,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:36,0.6.3
+cross_entropy,huggingface,no-grad-forward,speed,ms,V,vocab size,32768,2.6332000494003296,2.625375986099243,2.6437439918518066,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:36,0.6.3
+cross_entropy,huggingface,no-grad-forward,speed,ms,V,vocab size,65536,5.5122880935668945,5.5096830368042,5.516467094421387,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:36,0.6.3
+cross_entropy,huggingface,no-grad-forward,speed,ms,V,vocab size,131072,11.09065580368042,11.08307228088379,11.092691421508789,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:36,0.6.3
+cross_entropy,liger,full,memory,MB,V,vocab size,4096,256.32861328125,256.32861328125,256.32861328125,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:37,0.6.3
+cross_entropy,liger,full,memory,MB,V,vocab size,8192,512.32861328125,512.32861328125,512.32861328125,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:37,0.6.3
+cross_entropy,liger,full,memory,MB,V,vocab size,16384,1024.32861328125,1024.32861328125,1024.32861328125,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:37,0.6.3
+cross_entropy,liger,full,memory,MB,V,vocab size,32768,2048.32861328125,2048.32861328125,2048.32861328125,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:37,0.6.3
+cross_entropy,liger,full,memory,MB,V,vocab size,65536,4096.32861328125,4096.32861328125,4096.32861328125,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:37,0.6.3
+cross_entropy,liger,full,memory,MB,V,vocab size,131072,8192.328125,8192.328125,8192.328125,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:37,0.6.3
+cross_entropy,huggingface,full,memory,MB,V,vocab size,4096,1280.1259765625,1280.1259765625,1280.1259765625,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:37,0.6.3
+cross_entropy,huggingface,full,memory,MB,V,vocab size,8192,2560.1259765625,2560.1259765625,2560.1259765625,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:37,0.6.3
+cross_entropy,huggingface,full,memory,MB,V,vocab size,16384,5120.1259765625,5120.1259765625,5120.1259765625,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:37,0.6.3
+cross_entropy,huggingface,full,memory,MB,V,vocab size,32768,10240.1259765625,10240.1259765625,10240.1259765625,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:37,0.6.3
+cross_entropy,huggingface,full,memory,MB,V,vocab size,65536,20480.125,20480.125,20480.125,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:37,0.6.3
+cross_entropy,huggingface,full,memory,MB,V,vocab size,131072,40960.125,40960.125,40960.125,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:15:37,0.6.3
+distill_cosine_loss,liger,forward,speed,ms,BT,B x T,1024,5.482336044311523,5.424863815307617,5.4852800369262695,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:16:40,0.6.3
+distill_cosine_loss,liger,forward,speed,ms,BT,B x T,2048,11.046527862548828,10.958438301086424,11.086150360107421,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:16:40,0.6.3
+distill_cosine_loss,liger,forward,speed,ms,BT,B x T,4096,22.06049633026123,22.058547592163087,22.149740600585936,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:16:40,0.6.3
+distill_cosine_loss,liger,forward,speed,ms,BT,B x T,8192,43.879648208618164,43.68042831420899,44.07886810302734,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:16:40,0.6.3
+distill_cosine_loss,torch,forward,speed,ms,BT,B x T,1024,7.504976034164429,7.492262554168701,7.51921911239624,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:17:19,0.6.3
+distill_cosine_loss,torch,forward,speed,ms,BT,B x T,2048,14.59880018234253,14.595071792602539,14.602560043334961,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:17:19,0.6.3
+distill_cosine_loss,torch,forward,speed,ms,BT,B x T,4096,28.996320724487305,28.963756561279297,28.997856521606444,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:17:19,0.6.3
+distill_cosine_loss,torch,forward,speed,ms,BT,B x T,8192,57.974945068359375,57.974945068359375,57.974945068359375,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:17:19,0.6.3
+distill_cosine_loss,liger,full,speed,ms,BT,B x T,1024,7.4614081382751465,7.448761653900147,7.479840183258057,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:18:01,0.6.3
+distill_cosine_loss,liger,full,speed,ms,BT,B x T,2048,12.848031997680664,12.842541122436524,12.984524536132813,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:18:01,0.6.3
+distill_cosine_loss,liger,full,speed,ms,BT,B x T,4096,24.273263931274414,23.886354827880858,24.650790405273437,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:18:01,0.6.3
+distill_cosine_loss,liger,full,speed,ms,BT,B x T,8192,45.719425201416016,45.52298965454102,45.91586074829101,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:18:01,0.6.3
+distill_cosine_loss,torch,full,speed,ms,BT,B x T,1024,20.13145637512207,20.11712646484375,20.133882141113283,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:18:45,0.6.3
+distill_cosine_loss,torch,full,speed,ms,BT,B x T,2048,39.040449142456055,39.03628234863281,39.04461593627929,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:18:45,0.6.3
+distill_cosine_loss,torch,full,speed,ms,BT,B x T,4096,77.45779418945312,77.45779418945312,77.45779418945312,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:18:45,0.6.3
+distill_cosine_loss,torch,full,speed,ms,BT,B x T,8192,155.11053466796875,155.11053466796875,155.11053466796875,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:18:45,0.6.3
+distill_cosine_loss,liger,full,memory,MB,BT,B x T,1024,5107.01806640625,5107.01806640625,5107.01806640625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:19:23,0.6.3
+distill_cosine_loss,liger,full,memory,MB,BT,B x T,2048,5135.02587890625,5135.02587890625,5135.02587890625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:19:23,0.6.3
+distill_cosine_loss,liger,full,memory,MB,BT,B x T,4096,5191.04150390625,5191.04150390625,5191.04150390625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:19:23,0.6.3
+distill_cosine_loss,liger,full,memory,MB,BT,B x T,8192,5303.07275390625,5303.07275390625,5303.07275390625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:19:23,0.6.3
+distill_cosine_loss,torch,full,memory,MB,BT,B x T,1024,7614.0322265625,7614.0322265625,7614.0322265625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:20:03,0.6.3
+distill_cosine_loss,torch,full,memory,MB,BT,B x T,2048,11638.0634765625,11638.0634765625,11638.0634765625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:20:03,0.6.3
+distill_cosine_loss,torch,full,memory,MB,BT,B x T,4096,19702.125,19702.125,19702.125,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:20:03,0.6.3
+distill_cosine_loss,torch,full,memory,MB,BT,B x T,8192,35830.25,35830.25,35830.25,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:20:03,0.6.3
+distill_jsd_loss,liger,forward,speed,ms,BT,B x T,1024,6.150176048278809,6.088435077667237,6.210643291473389,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:21:04,0.6.3
+distill_jsd_loss,liger,forward,speed,ms,BT,B x T,2048,12.295968055725098,12.259296035766603,12.413913917541503,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:21:04,0.6.3
+distill_jsd_loss,liger,forward,speed,ms,BT,B x T,4096,24.43889617919922,24.377970886230468,24.475283813476565,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:21:04,0.6.3
+distill_jsd_loss,liger,forward,speed,ms,BT,B x T,8192,49.06742477416992,48.95343475341797,49.18141479492188,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:21:04,0.6.3
+distill_jsd_loss,torch,forward,speed,ms,BT,B x T,1024,15.150000095367432,15.14681625366211,15.157792091369629,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:21:42,0.6.3
+distill_jsd_loss,torch,forward,speed,ms,BT,B x T,2048,29.87424087524414,29.866349029541016,29.876505661010743,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:21:42,0.6.3
+distill_jsd_loss,torch,forward,speed,ms,BT,B x T,4096,62.76121520996094,62.76121520996094,62.76121520996094,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:21:42,0.6.3
+distill_jsd_loss,torch,forward,speed,ms,BT,B x T,8192,125.51312255859375,125.51312255859375,125.51312255859375,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:21:42,0.6.3
+distill_jsd_loss,liger,backward,speed,ms,BT,B x T,1024,0.5792319774627686,0.5751296043395996,0.5889855861663819,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:22:18,0.6.3
+distill_jsd_loss,liger,backward,speed,ms,BT,B x T,2048,0.5791360139846802,0.5747583746910095,0.5872704148292541,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:22:18,0.6.3
+distill_jsd_loss,liger,backward,speed,ms,BT,B x T,4096,0.5834559798240662,0.578655982017517,0.5926912069320679,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:22:18,0.6.3
+distill_jsd_loss,liger,backward,speed,ms,BT,B x T,8192,0.5930559933185577,0.5893375873565674,0.6031744003295898,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:22:18,0.6.3
+distill_jsd_loss,torch,backward,speed,ms,BT,B x T,1024,11.423744201660156,11.419501113891602,11.425446319580079,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:22:55,0.6.3
+distill_jsd_loss,torch,backward,speed,ms,BT,B x T,2048,22.10854434967041,22.107513427734375,22.11232719421387,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:22:55,0.6.3
+distill_jsd_loss,torch,backward,speed,ms,BT,B x T,4096,42.777984619140625,42.777677917480474,42.77829132080078,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:22:55,0.6.3
+distill_jsd_loss,torch,backward,speed,ms,BT,B x T,8192,85.08537292480469,85.08537292480469,85.08537292480469,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:22:55,0.6.3
+distill_jsd_loss,liger,full,speed,ms,BT,B x T,1024,16.75990390777588,9.177663803100586,21.292064666748047,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:23:34,0.6.3
+distill_jsd_loss,liger,full,speed,ms,BT,B x T,2048,13.954479694366455,13.876607894897461,13.97532844543457,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:23:34,0.6.3
+distill_jsd_loss,liger,full,speed,ms,BT,B x T,4096,26.2673282623291,26.174898910522458,26.29468765258789,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:23:34,0.6.3
+distill_jsd_loss,liger,full,speed,ms,BT,B x T,8192,51.13139343261719,51.13139343261719,51.13139343261719,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:23:34,0.6.3
+distill_jsd_loss,torch,full,speed,ms,BT,B x T,1024,26.59667205810547,26.596134185791016,26.598841857910156,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:24:12,0.6.3
+distill_jsd_loss,torch,full,speed,ms,BT,B x T,2048,52.00624084472656,52.00624084472656,52.00624084472656,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:24:12,0.6.3
+distill_jsd_loss,torch,full,speed,ms,BT,B x T,4096,105.69990539550781,105.69990539550781,105.69990539550781,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:24:12,0.6.3
+distill_jsd_loss,torch,full,speed,ms,BT,B x T,8192,210.88748168945312,210.88748168945312,210.88748168945312,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:24:12,0.6.3
+distill_jsd_loss,liger,full,memory,MB,BT,B x T,1024,5350.53955078125,5350.53955078125,5350.53955078125,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:24:46,0.6.3
+distill_jsd_loss,liger,full,memory,MB,BT,B x T,2048,5378.54736328125,5378.54736328125,5378.54736328125,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:24:46,0.6.3
+distill_jsd_loss,liger,full,memory,MB,BT,B x T,4096,5434.56298828125,5434.56298828125,5434.56298828125,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:24:46,0.6.3
+distill_jsd_loss,liger,full,memory,MB,BT,B x T,8192,5546.59423828125,5546.59423828125,5546.59423828125,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:24:46,0.6.3
+distill_jsd_loss,torch,full,memory,MB,BT,B x T,1024,10110.009765625,10110.009765625,10110.009765625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:25:24,0.6.3
+distill_jsd_loss,torch,full,memory,MB,BT,B x T,2048,16647.017578125,16647.017578125,16647.017578125,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:25:24,0.6.3
+distill_jsd_loss,torch,full,memory,MB,BT,B x T,4096,29721.033203125,29721.033203125,29721.033203125,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:25:24,0.6.3
+distill_jsd_loss,torch,full,memory,MB,BT,B x T,8192,55869.0625,55869.0625,55869.0625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": false, ""weight_hard_loss"": 0.5, ""weight_soft_loss"": 0.5, ""ignore_index"": -100}",NVIDIA H100 80GB HBM3,2025-11-21 00:25:24,0.6.3
+dpo_loss,liger,forward,speed,ms,B,Batch Size (B),2,3.460911989212036,3.4570751667022703,3.4746176242828373,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:25:59,0.6.3
+dpo_loss,liger,forward,speed,ms,B,Batch Size (B),4,6.803647994995117,6.799212837219239,6.855993556976318,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:25:59,0.6.3
+dpo_loss,liger,forward,speed,ms,B,Batch Size (B),8,13.362688064575195,13.360921478271484,13.37086067199707,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:25:59,0.6.3
+dpo_loss,liger,forward,speed,ms,B,Batch Size (B),16,26.97711944580078,26.84821090698242,27.050214004516604,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:25:59,0.6.3
+dpo_loss,liger,forward,speed,ms,B,Batch Size (B),32,53.92243194580078,53.92243194580078,53.92243194580078,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:25:59,0.6.3
+dpo_loss,huggingface,forward,speed,ms,B,Batch Size (B),2,2.9045599699020386,2.8993088245391845,2.9089407444000246,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:26:14,0.6.3
+dpo_loss,huggingface,forward,speed,ms,B,Batch Size (B),4,5.488831996917725,5.485017585754394,5.494470405578613,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:26:14,0.6.3
+dpo_loss,huggingface,forward,speed,ms,B,Batch Size (B),8,10.713248252868652,10.707571029663086,10.718950080871583,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:26:14,0.6.3
+dpo_loss,huggingface,forward,speed,ms,B,Batch Size (B),16,20.915871620178223,20.91204490661621,20.938719177246092,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:26:14,0.6.3
+dpo_loss,huggingface,forward,speed,ms,B,Batch Size (B),32,41.67644691467285,41.673509216308595,41.679384613037115,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:26:14,0.6.3
+dpo_loss,liger,backward,speed,ms,B,Batch Size (B),2,0.3117760121822357,0.3084991991519928,0.31986559629440303,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:26:29,0.6.3
+dpo_loss,liger,backward,speed,ms,B,Batch Size (B),4,0.3109759986400604,0.3084415912628174,0.31614080667495725,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:26:29,0.6.3
+dpo_loss,liger,backward,speed,ms,B,Batch Size (B),8,0.3119039982557297,0.30905601382255554,0.31782400608062744,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:26:29,0.6.3
+dpo_loss,liger,backward,speed,ms,B,Batch Size (B),16,0.3120959997177124,0.3086976110935211,0.3193664014339447,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:26:29,0.6.3
+dpo_loss,liger,backward,speed,ms,B,Batch Size (B),32,0.31171199679374695,0.30785279273986815,0.3202816009521484,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:26:29,0.6.3
+dpo_loss,huggingface,backward,speed,ms,B,Batch Size (B),2,1.8959360122680664,1.8944447994232179,1.8975808143615722,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:26:44,0.6.3
+dpo_loss,huggingface,backward,speed,ms,B,Batch Size (B),4,3.3886719942092896,3.3878719806671143,3.390515232086182,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:26:44,0.6.3
+dpo_loss,huggingface,backward,speed,ms,B,Batch Size (B),8,6.372064113616943,6.365964889526367,6.375712203979492,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:26:44,0.6.3
+dpo_loss,huggingface,backward,speed,ms,B,Batch Size (B),16,12.337823867797852,12.33246078491211,12.342809867858888,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:26:44,0.6.3
+dpo_loss,huggingface,backward,speed,ms,B,Batch Size (B),32,24.10171127319336,24.09772071838379,24.107865524291995,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:26:44,0.6.3
+dpo_loss,liger,full,speed,ms,B,Batch Size (B),2,5.656288146972656,5.590713596343994,5.813004684448242,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:27:00,0.6.3
+dpo_loss,liger,full,speed,ms,B,Batch Size (B),4,8.925344467163086,8.720000267028809,8.941439628601074,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:27:00,0.6.3
+dpo_loss,liger,full,speed,ms,B,Batch Size (B),8,16.282544136047363,15.819711685180664,16.506912231445312,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:27:00,0.6.3
+dpo_loss,liger,full,speed,ms,B,Batch Size (B),16,29.31705665588379,29.285107040405272,29.389363861083986,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:27:00,0.6.3
+dpo_loss,liger,full,speed,ms,B,Batch Size (B),32,57.39100646972656,57.39100646972656,57.39100646972656,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:27:00,0.6.3
+dpo_loss,huggingface,full,speed,ms,B,Batch Size (B),2,4.810223817825317,4.802495956420898,4.815552043914795,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:27:16,0.6.3
+dpo_loss,huggingface,full,speed,ms,B,Batch Size (B),4,8.894623756408691,8.89020824432373,8.896736145019531,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:27:16,0.6.3
+dpo_loss,huggingface,full,speed,ms,B,Batch Size (B),8,17.104415893554688,17.08787269592285,17.118854904174803,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:27:16,0.6.3
+dpo_loss,huggingface,full,speed,ms,B,Batch Size (B),16,33.28947067260742,33.28588180541992,33.29260177612305,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:27:16,0.6.3
+dpo_loss,huggingface,full,speed,ms,B,Batch Size (B),32,66.01538848876953,66.01538848876953,66.01538848876953,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:27:16,0.6.3
+dpo_loss,liger,full,memory,MB,B,Batch Size (B),2,2829.50634765625,2829.50634765625,2829.50634765625,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:27:32,0.6.3
+dpo_loss,liger,full,memory,MB,B,Batch Size (B),4,2835.515625,2835.515625,2835.515625,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:27:32,0.6.3
+dpo_loss,liger,full,memory,MB,B,Batch Size (B),8,2847.5341796875,2847.5341796875,2847.5341796875,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:27:32,0.6.3
+dpo_loss,liger,full,memory,MB,B,Batch Size (B),16,2871.5712890625,2871.5712890625,2871.5712890625,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:27:32,0.6.3
+dpo_loss,liger,full,memory,MB,B,Batch Size (B),32,2919.6455078125,2919.6455078125,2919.6455078125,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:27:32,0.6.3
+dpo_loss,huggingface,full,memory,MB,B,Batch Size (B),2,2827.24267578125,2827.24267578125,2827.24267578125,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:27:48,0.6.3
+dpo_loss,huggingface,full,memory,MB,B,Batch Size (B),4,4333.2900390625,4333.2900390625,4333.2900390625,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:27:48,0.6.3
+dpo_loss,huggingface,full,memory,MB,B,Batch Size (B),8,7345.3447265625,7345.3447265625,7345.3447265625,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:27:48,0.6.3
+dpo_loss,huggingface,full,memory,MB,B,Batch Size (B),16,13373.4541015625,13373.4541015625,13373.4541015625,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:27:48,0.6.3
+dpo_loss,huggingface,full,memory,MB,B,Batch Size (B),32,25429.671875,25429.671875,25429.671875,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:27:48,0.6.3
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,1024,0.012095999903976917,0.011872000060975552,0.012319999746978283,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:07,0.6.3
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,2048,0.017791999503970146,0.01756799966096878,0.018079999834299088,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:07,0.6.3
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,3072,0.023264000192284584,0.02300800010561943,0.02364799939095974,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:07,0.6.3
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,4096,0.029472000896930695,0.029184000566601753,0.030208000913262367,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:07,0.6.3
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,5120,0.036896001547575,0.03656959906220436,0.03750399872660637,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:07,0.6.3
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,6144,0.040800001472234726,0.04044799879193306,0.0416640006005764,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:07,0.6.3
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,7168,0.04816000163555145,0.047839999198913574,0.048895999789237976,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:07,0.6.3
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,8192,0.05183999985456467,0.051392000168561935,0.052671998739242554,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:07,0.6.3
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,9216,0.05967999994754791,0.059039998799562454,0.06015999987721443,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:07,0.6.3
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,10240,0.06332799792289734,0.062431998550891876,0.06371200084686279,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:07,0.6.3
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,11264,0.070592001080513,0.06992000341415405,0.07107199728488922,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:07,0.6.3
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,12288,0.07436800003051758,0.07347200065851212,0.07472000271081924,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:07,0.6.3
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,13312,0.08158400282263756,0.08073599636554718,0.08188799768686295,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:07,0.6.3
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,14336,0.08524800091981888,0.08464639633893967,0.08562559783458709,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:07,0.6.3
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,15360,0.09270399808883667,0.09228800237178802,0.09302400052547455,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:07,0.6.3
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,16384,0.09651199728250504,0.09625600278377533,0.09678079932928085,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:07,0.6.3
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,1024,0.0480320006608963,0.04771199822425842,0.04854400083422661,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:15,0.6.3
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,2048,0.09593600034713745,0.095551997423172,0.09628800302743912,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:15,0.6.3
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,3072,0.13631999492645264,0.13583999872207642,0.1369280070066452,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:15,0.6.3
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,4096,0.17839999496936798,0.177824005484581,0.17878399789333344,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:15,0.6.3
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,5120,0.218176007270813,0.2178560048341751,0.21862399578094482,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:15,0.6.3
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,6144,0.25839999318122864,0.25763198733329773,0.2590720057487488,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:15,0.6.3
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,7168,0.2985599935054779,0.2980479896068573,0.29897600412368774,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:15,0.6.3
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,8192,0.33612799644470215,0.3357760012149811,0.33667200803756714,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:15,0.6.3
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,9216,0.3790079951286316,0.3783999979496002,0.3794560134410858,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:15,0.6.3
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,10240,0.41862401366233826,0.4182080030441284,0.4190720021724701,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:15,0.6.3
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,11264,0.45686399936676025,0.4562560021877289,0.4574719965457916,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:15,0.6.3
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,12288,0.49686399102211,0.49644801020622253,0.4973120093345642,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:15,0.6.3
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,13312,0.5390719771385193,0.5386175870895387,0.5398079752922058,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:15,0.6.3
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,14336,0.5771840214729309,0.5764480233192444,0.5776960253715515,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:15,0.6.3
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,15360,0.6190720200538635,0.6186559796333313,0.6194559931755066,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:15,0.6.3
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,16384,0.6593919992446899,0.6587200164794922,0.6599040031433105,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:15,0.6.3
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,1024,0.04419200122356415,0.04105599969625473,0.057792000472545624,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:30,0.6.3
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,2048,0.043168000876903534,0.04012800008058548,0.056352000683546066,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:30,0.6.3
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,3072,0.04249599948525429,0.039744000881910324,0.05137280002236366,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:30,0.6.3
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,4096,0.044415999203920364,0.04214400053024292,0.05801599845290184,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:30,0.6.3
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,5120,0.04931199923157692,0.04825599864125252,0.052563201636075974,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:30,0.6.3
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,6144,0.056832000613212585,0.056095998734235764,0.05756799876689911,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:30,0.6.3
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,7168,0.0655680000782013,0.06460800021886826,0.06601600348949432,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:30,0.6.3
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,8192,0.07379200309515,0.07286400347948074,0.07420799881219864,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:30,0.6.3
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,9216,0.3769280016422272,0.3763520121574402,0.377375990152359,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:30,0.6.3
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,10240,0.4164479970932007,0.416128009557724,0.41676801443099976,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:30,0.6.3
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,11264,0.4567039906978607,0.45612800121307373,0.4572800099849701,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:30,0.6.3
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,12288,0.4986560046672821,0.49833598732948303,0.4990079998970032,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:30,0.6.3
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,13312,0.5388479828834534,0.5384320020675659,0.5395519733428955,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:30,0.6.3
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,14336,0.579039990901947,0.5783360004425049,0.5795199871063232,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:30,0.6.3
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,15360,0.6163520216941833,0.6159999966621399,0.6168000102043152,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:30,0.6.3
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,16384,0.6570240259170532,0.6564480066299438,0.6575807809829712,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:30,0.6.3
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,1024,0.24964799731969833,0.2208319991827011,0.3375744044780732,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:41,0.6.3
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,2048,0.2091040015220642,0.19392000138759613,0.2533760070800781,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:41,0.6.3
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,3072,0.2088640034198761,0.18991999328136444,0.25528960227966313,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:41,0.6.3
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,4096,0.2120319977402687,0.19366399943828583,0.26048640012741087,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:41,0.6.3
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,5120,0.21542400121688843,0.19551999866962433,0.2648832023143769,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:41,0.6.3
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,6144,0.21723199635744095,0.19898239672183993,0.26967679858207716,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:41,0.6.3
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,7168,0.21171200275421143,0.1940159946680069,0.2571839988231659,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:41,0.6.3
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,8192,0.21967999637126923,0.19921280145645143,0.2733440101146698,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:41,0.6.3
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,9216,0.21184000372886658,0.19685760140419006,0.2564736008644105,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:41,0.6.3
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,10240,0.21302400529384613,0.1987839937210083,0.2625407874584198,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:41,0.6.3
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,11264,0.2149920016527176,0.20200959742069244,0.2646143972873688,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:41,0.6.3
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,12288,0.22286400198936462,0.20885760188102723,0.26769920587539675,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:41,0.6.3
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,13312,0.22430399805307388,0.2012800008058548,0.28225278854370117,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:41,0.6.3
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,14336,0.22572800517082214,0.2054464012384415,0.28650879859924316,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:41,0.6.3
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,15360,0.21347199380397797,0.19682559967041016,0.2638272106647491,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:41,0.6.3
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,16384,0.20684799551963806,0.20188799500465393,0.24622720181941987,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:41,0.6.3
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,1024,0.14972800016403198,0.1483840048313141,0.17528319358825684,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:49,0.6.3
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,2048,0.2727360129356384,0.2719680070877075,0.27350398898124695,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:49,0.6.3
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,3072,0.38649600744247437,0.3856639862060547,0.38729599118232727,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:49,0.6.3
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,4096,0.5001919865608215,0.4991680085659027,0.5008959770202637,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:49,0.6.3
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,5120,0.6049280166625977,0.604095995426178,0.6055999994277954,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:49,0.6.3
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,6144,0.714847981929779,0.7140160202980042,0.71561598777771,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:49,0.6.3
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,7168,0.8189759850502014,0.8179519772529602,0.8198400139808655,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:49,0.6.3
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,8192,0.9297279715538025,0.9287999868392944,0.9306560158729553,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:49,0.6.3
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,9216,1.0325599908828735,1.0315776109695434,1.033510398864746,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:49,0.6.3
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,10240,1.139792025089264,1.1387519836425781,1.140768051147461,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:49,0.6.3
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,11264,1.2446080446243286,1.24343683719635,1.2458239793777466,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:49,0.6.3
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,12288,1.3549920320510864,1.3536640405654907,1.3561919927597046,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:49,0.6.3
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,13312,1.4573440551757812,1.456268811225891,1.4581760168075562,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:49,0.6.3
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,14336,1.5654560327529907,1.5645504236221313,1.566540789604187,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:49,0.6.3
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,15360,1.6701599955558777,1.6690752267837525,1.6711807966232302,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:49,0.6.3
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,16384,1.7799359560012817,1.7786240577697754,1.7807040214538574,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:28:49,0.6.3
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,1024,0.23401600122451782,0.21018240451812745,0.29951359033584596,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:03,0.6.3
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,2048,0.24057599902153015,0.21266559958457948,0.30333440303802495,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:03,0.6.3
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,3072,0.22617599368095398,0.20549760460853578,0.2815551996231079,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:03,0.6.3
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,4096,0.23656000196933746,0.20854400098323822,0.30458880662918103,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:03,0.6.3
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,5120,0.2287999987602234,0.204652801156044,0.30003198981285095,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:03,0.6.3
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,6144,0.2311680018901825,0.2292799949645996,0.26945280432701113,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:03,0.6.3
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,7168,0.2574079930782318,0.2563520073890686,0.2590399980545044,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:03,0.6.3
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,8192,0.289792001247406,0.28915199637413025,0.29052799940109253,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:03,0.6.3
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,9216,1.032256007194519,1.0311296224594115,1.0330943822860719,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:03,0.6.3
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,10240,1.1396479606628418,1.1386560201644897,1.1404608011245727,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:03,0.6.3
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,11264,1.2446399927139282,1.2434687852859496,1.2459839582443237,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:03,0.6.3
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,12288,1.356160044670105,1.35481595993042,1.3574719429016113,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:03,0.6.3
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,13312,1.4568320512771606,1.4557119607925415,1.4578880071640015,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:03,0.6.3
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,14336,1.5653120279312134,1.564339208602905,1.56638081073761,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:03,0.6.3
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,15360,1.6699360013008118,1.6687359809875488,1.67084801197052,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:03,0.6.3
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,16384,1.779968023300171,1.7787519693374634,1.7811199426651,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:03,0.6.3
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,1024,0.33452799916267395,0.3110976099967957,0.3890303969383241,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:11,0.6.3
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,2048,0.3325119912624359,0.30963200330734253,0.377375990152359,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:11,0.6.3
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,3072,0.3339200019836426,0.3097215950489044,0.3864256083965302,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:11,0.6.3
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,4096,0.3298240005970001,0.3080959916114807,0.3831680119037628,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:11,0.6.3
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,5120,0.33347201347351074,0.3109248042106628,0.37084798812866215,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:11,0.6.3
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,6144,0.33500799536705017,0.313676792383194,0.38284800052642826,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:11,0.6.3
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,7168,0.3349120020866394,0.31004160046577456,0.38584319949150087,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:11,0.6.3
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,8192,0.32529599964618683,0.3040895998477936,0.3655103981494904,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:11,0.6.3
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,9216,0.33692799508571625,0.31200640797615053,0.38866558670997625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:11,0.6.3
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,10240,0.3361919969320297,0.31061760187149046,0.3869184017181396,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:11,0.6.3
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,11264,0.3304319977760315,0.30742400884628296,0.38073599338531494,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:11,0.6.3
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,12288,0.3357919901609421,0.3141823947429657,0.3881343901157379,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:11,0.6.3
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,13312,0.34564799070358276,0.32259199023246765,0.39459840655326844,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:11,0.6.3
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,14336,0.34272000193595886,0.31900799870491026,0.39977599978446965,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:11,0.6.3
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,15360,0.34624001383781433,0.32325119972229,0.3974847912788391,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:11,0.6.3
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,16384,0.34508800506591797,0.3202367901802063,0.4026048064231873,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:11,0.6.3
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,1024,0.200095996260643,0.19353599846363068,0.2512640058994293,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:19,0.6.3
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,2048,0.3643519878387451,0.3624640107154846,0.3667519986629486,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:19,0.6.3
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,3072,0.5226560235023499,0.5208832144737243,0.5252607941627503,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:19,0.6.3
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,4096,0.6723840236663818,0.6705536007881165,0.6750144004821778,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:19,0.6.3
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,5120,0.8215039968490601,0.819539201259613,0.823852825164795,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:19,0.6.3
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,6144,0.9688320159912109,0.9667775988578796,0.9714303851127625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:19,0.6.3
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,7168,1.116047978401184,1.114348793029785,1.1185791969299317,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:19,0.6.3
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,8192,1.2627520561218262,1.260633611679077,1.2650879621505737,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:19,0.6.3
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,9216,1.4095360040664673,1.4078976154327392,1.4119999408721924,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:19,0.6.3
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,10240,1.5544960498809814,1.552672028541565,1.5571839809417725,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:19,0.6.3
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,11264,1.7024959921836853,1.7004159688949585,1.7055039405822754,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:19,0.6.3
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,12288,1.851680040359497,1.8494080305099487,1.8541439771652222,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:19,0.6.3
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,13312,1.996448040008545,1.9945791721343995,1.9990719556808472,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:19,0.6.3
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,14336,2.142848014831543,2.1408191680908204,2.1458175659179686,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:19,0.6.3
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,15360,2.289583921432495,2.287513542175293,2.291654348373413,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:19,0.6.3
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,16384,2.4368960857391357,2.434777545928955,2.438892889022827,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:19,0.6.3
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,1024,0.44806399941444397,0.4219839870929718,0.5084224104881286,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,2048,0.45737600326538086,0.4270080029964447,0.5459840297698975,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,3072,0.4562720060348511,0.427347195148468,0.5161920070648194,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,4096,0.46854400634765625,0.43539839386940005,0.5673280000686646,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,5120,0.48528000712394714,0.450739198923111,0.6034367918968202,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,6144,0.4816319942474365,0.44728959202766416,0.6066367864608765,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,7168,0.4683679938316345,0.4354752063751221,0.5749952077865601,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,8192,0.4551040083169937,0.4264639914035797,0.5004352092742921,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,9216,1.4110080003738403,1.408959984779358,1.413375973701477,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,10240,1.5563839673995972,1.5544320344924927,1.5588608264923094,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,11264,1.703536033630371,1.7015680074691772,1.7061439752578735,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,12288,1.8510719537734985,1.8491455793380738,1.8538944244384767,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,13312,1.9967679977416992,1.9948480129241943,1.999078392982483,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,14336,2.1429760456085205,2.1412160396575928,2.145785617828369,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,15360,2.2898240089416504,2.28786563873291,2.292185640335083,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,16384,2.4369280338287354,2.4348158836364746,2.4395328044891356,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,liger,full,memory,MB,hidden_size,hidden_size,1024,40.53857421875,40.53857421875,40.53857421875,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,liger,full,memory,MB,hidden_size,hidden_size,2048,81.07470703125,81.07470703125,81.07470703125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,liger,full,memory,MB,hidden_size,hidden_size,3072,121.61083984375,121.61083984375,121.61083984375,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,liger,full,memory,MB,hidden_size,hidden_size,4096,162.14697265625,162.14697265625,162.14697265625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,liger,full,memory,MB,hidden_size,hidden_size,5120,202.68310546875,202.68310546875,202.68310546875,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,liger,full,memory,MB,hidden_size,hidden_size,6144,243.21923828125,243.21923828125,243.21923828125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,liger,full,memory,MB,hidden_size,hidden_size,7168,283.75537109375,283.75537109375,283.75537109375,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,liger,full,memory,MB,hidden_size,hidden_size,8192,324.29150390625,324.29150390625,324.29150390625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,liger,full,memory,MB,hidden_size,hidden_size,9216,364.82763671875,364.82763671875,364.82763671875,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,liger,full,memory,MB,hidden_size,hidden_size,10240,405.36376953125,405.36376953125,405.36376953125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,liger,full,memory,MB,hidden_size,hidden_size,11264,445.89990234375,445.89990234375,445.89990234375,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,liger,full,memory,MB,hidden_size,hidden_size,12288,486.43603515625,486.43603515625,486.43603515625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,liger,full,memory,MB,hidden_size,hidden_size,13312,526.97216796875,526.97216796875,526.97216796875,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,liger,full,memory,MB,hidden_size,hidden_size,14336,568.28955078125,568.28955078125,568.28955078125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,liger,full,memory,MB,hidden_size,hidden_size,15360,608.31005859375,608.31005859375,608.31005859375,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,liger,full,memory,MB,hidden_size,hidden_size,16384,648.58056640625,648.58056640625,648.58056640625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch,full,memory,MB,hidden_size,hidden_size,1024,136.02197265625,136.02197265625,136.02197265625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch,full,memory,MB,hidden_size,hidden_size,2048,272.04150390625,272.04150390625,272.04150390625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch,full,memory,MB,hidden_size,hidden_size,3072,408.06103515625,408.06103515625,408.06103515625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch,full,memory,MB,hidden_size,hidden_size,4096,544.08056640625,544.08056640625,544.08056640625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch,full,memory,MB,hidden_size,hidden_size,5120,652.60009765625,652.60009765625,652.60009765625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch,full,memory,MB,hidden_size,hidden_size,6144,768.095703125,768.095703125,768.095703125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch,full,memory,MB,hidden_size,hidden_size,7168,896.111328125,896.111328125,896.111328125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch,full,memory,MB,hidden_size,hidden_size,8192,1024.126953125,1024.126953125,1024.126953125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch,full,memory,MB,hidden_size,hidden_size,9216,1152.142578125,1152.142578125,1152.142578125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch,full,memory,MB,hidden_size,hidden_size,10240,1280.158203125,1280.158203125,1280.158203125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch,full,memory,MB,hidden_size,hidden_size,11264,1408.173828125,1408.173828125,1408.173828125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch,full,memory,MB,hidden_size,hidden_size,12288,1536.189453125,1536.189453125,1536.189453125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch,full,memory,MB,hidden_size,hidden_size,13312,1664.205078125,1664.205078125,1664.205078125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch,full,memory,MB,hidden_size,hidden_size,14336,1792.220703125,1792.220703125,1792.220703125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch,full,memory,MB,hidden_size,hidden_size,15360,1920.236328125,1920.236328125,1920.236328125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch,full,memory,MB,hidden_size,hidden_size,16384,2048.251953125,2048.251953125,2048.251953125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,memory,MB,hidden_size,hidden_size,1024,64.02392578125,64.02392578125,64.02392578125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,memory,MB,hidden_size,hidden_size,2048,128.04541015625,128.04541015625,128.04541015625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,memory,MB,hidden_size,hidden_size,3072,192.06298828125,192.06298828125,192.06298828125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,memory,MB,hidden_size,hidden_size,4096,256.08251953125,256.08251953125,256.08251953125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,memory,MB,hidden_size,hidden_size,5120,320.10205078125,320.10205078125,320.10205078125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,memory,MB,hidden_size,hidden_size,6144,384.12158203125,384.12158203125,384.12158203125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,memory,MB,hidden_size,hidden_size,7168,448.14111328125,448.14111328125,448.14111328125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,memory,MB,hidden_size,hidden_size,8192,512.16064453125,512.16064453125,512.16064453125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,memory,MB,hidden_size,hidden_size,9216,1152.142578125,1152.142578125,1152.142578125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,memory,MB,hidden_size,hidden_size,10240,1280.158203125,1280.158203125,1280.158203125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,memory,MB,hidden_size,hidden_size,11264,1408.173828125,1408.173828125,1408.173828125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,memory,MB,hidden_size,hidden_size,12288,1536.189453125,1536.189453125,1536.189453125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,memory,MB,hidden_size,hidden_size,13312,1664.205078125,1664.205078125,1664.205078125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,memory,MB,hidden_size,hidden_size,14336,1792.220703125,1792.220703125,1792.220703125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,memory,MB,hidden_size,hidden_size,15360,1920.236328125,1920.236328125,1920.236328125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=False,torch_compile,full,memory,MB,hidden_size,hidden_size,16384,2048.251953125,2048.251953125,2048.251953125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:28,0.6.3
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,1024,0.012864000163972378,0.012703999876976013,0.013055999763309956,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:39,0.6.3
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,2048,0.018880000337958336,0.018688000738620758,0.019168000668287277,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:39,0.6.3
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,3072,0.024191999807953835,0.024032000452280045,0.024607999250292778,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:39,0.6.3
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,4096,0.03110400028526783,0.030848000198602676,0.03174399957060814,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:39,0.6.3
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,5120,0.03766399994492531,0.037376001477241516,0.03827200084924698,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:39,0.6.3
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,6144,0.041760001331567764,0.0414079986512661,0.042527999728918076,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:39,0.6.3
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,7168,0.04918399825692177,0.04879999905824661,0.049855999648571014,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:39,0.6.3
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,8192,0.05337600037455559,0.05289600044488907,0.0541439987719059,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:39,0.6.3
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,9216,0.06043200008571148,0.05974400043487549,0.06092799827456474,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:39,0.6.3
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,10240,0.06451199948787689,0.06393600255250931,0.06489600241184235,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:39,0.6.3
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,11264,0.07209599763154984,0.07126399874687195,0.07238399982452393,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:39,0.6.3
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,12288,0.0761599987745285,0.07523199915885925,0.0764480009675026,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:39,0.6.3
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,13312,0.08355200290679932,0.08275199681520462,0.08380799740552902,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:39,0.6.3
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,14336,0.08774399757385254,0.08726400136947632,0.08799999952316284,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:39,0.6.3
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,15360,0.09491200000047684,0.09455999732017517,0.09523200243711472,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:39,0.6.3
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,16384,0.09913600236177444,0.09881599992513657,0.09942399710416794,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:39,0.6.3
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,1024,0.061503998935222626,0.05998719856142998,0.062111999839544296,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:48,0.6.3
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,2048,0.1265600025653839,0.12505599856376648,0.1271360069513321,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:48,0.6.3
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,3072,0.18236799538135529,0.18095999956130981,0.1831039935350418,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:48,0.6.3
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,4096,0.23580799996852875,0.2343679964542389,0.2364480048418045,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:48,0.6.3
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,5120,0.2895039916038513,0.2877439856529236,0.28998398780822754,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:48,0.6.3
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,6144,0.34278398752212524,0.3415679931640625,0.3437120020389557,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:48,0.6.3
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,7168,0.3965120017528534,0.39481601119041443,0.3970560133457184,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:48,0.6.3
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,8192,0.44993600249290466,0.4484800100326538,0.4505920112133026,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:48,0.6.3
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,9216,0.5036799907684326,0.5018879771232605,0.5041279792785645,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:48,0.6.3
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,10240,0.5568959712982178,0.5554559826850891,0.5576959848403931,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:48,0.6.3
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,11264,0.6104959845542908,0.6088576078414917,0.610975980758667,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:48,0.6.3
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,12288,0.6639999747276306,0.6627391934394836,0.6648447990417481,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:48,0.6.3
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,13312,0.7177600264549255,0.7162240147590637,0.7183039784431458,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:48,0.6.3
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,14336,0.7714239954948425,0.7700480222702026,0.7722880244255066,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:48,0.6.3
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,15360,0.824895977973938,0.8233536243438722,0.8254463911056519,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:48,0.6.3
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,16384,0.8785920143127441,0.8771520256996155,0.8791679739952087,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:48,0.6.3
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,1024,0.06159999966621399,0.05993599817156792,0.06220800057053566,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:56,0.6.3
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,2048,0.12665599584579468,0.12511999905109406,0.12729600071907043,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:56,0.6.3
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,3072,0.18243199586868286,0.18083840012550356,0.1831039935350418,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:56,0.6.3
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,4096,0.2359360009431839,0.23443199694156647,0.23657600581645966,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:56,0.6.3
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,5120,0.28942400217056274,0.2876479923725128,0.28992000222206116,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:56,0.6.3
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,6144,0.3429119884967804,0.34166398644447327,0.34380799531936646,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:56,0.6.3
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,7168,0.3967039883136749,0.3949120044708252,0.39715200662612915,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:56,0.6.3
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,8192,0.45001599192619324,0.4485119879245758,0.450655996799469,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:56,0.6.3
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,9216,0.5035520195960999,0.5018560290336609,0.5040640234947205,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:56,0.6.3
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,10240,0.5570560097694397,0.5555776119232178,0.5578240156173706,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:56,0.6.3
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,11264,0.6105599999427795,0.6088512063026428,0.6110079884529114,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:56,0.6.3
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,12288,0.6640959978103638,0.6627328038215637,0.6649919748306274,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:56,0.6.3
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,13312,0.7176960110664368,0.7160767793655396,0.7182400226593018,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:56,0.6.3
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,14336,0.7715200185775757,0.7699519991874695,0.7722880244255066,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:56,0.6.3
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,15360,0.8248639702796936,0.8233280181884766,0.8254143834114075,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:56,0.6.3
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,16384,0.8785600066184998,0.8768960237503052,0.8791679739952087,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:29:56,0.6.3
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,1024,0.2593280076980591,0.2353407979011536,0.325324809551239,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:07,0.6.3
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,2048,0.2680639922618866,0.2437887966632843,0.33909120559692396,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:07,0.6.3
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,3072,0.25809600949287415,0.23511040210723877,0.33762560486793525,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:07,0.6.3
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,4096,0.2627519965171814,0.2409023940563202,0.33111039996147157,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:07,0.6.3
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,5120,0.26598401367664337,0.24250240325927735,0.34199680685997014,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:07,0.6.3
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,6144,0.2672799974679947,0.24316799640655518,0.3343679904937744,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:07,0.6.3
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,7168,0.2719839960336685,0.24771200120449066,0.3326848089694977,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:07,0.6.3
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,8192,0.26924799382686615,0.24713599681854248,0.3442240059375763,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:07,0.6.3
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,9216,0.25964799523353577,0.23783680200576782,0.33433598875999454,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:07,0.6.3
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,10240,0.2659519910812378,0.24266240596771238,0.3404608011245728,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:07,0.6.3
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,11264,0.26684799790382385,0.24304640293121338,0.3321727991104126,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:07,0.6.3
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,12288,0.2671999931335449,0.2452544033527374,0.33614079356193544,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:07,0.6.3
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,13312,0.28993599116802216,0.2579648017883301,0.3606016099452973,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:07,0.6.3
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,14336,0.3184640109539032,0.2858624041080475,0.48444159626960764,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:07,0.6.3
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,15360,0.2801920026540756,0.2555840015411377,0.33638399839401245,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:07,0.6.3
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,16384,0.29155199229717255,0.2628928005695343,0.33980159163475043,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:07,0.6.3
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,1024,0.17945599555969238,0.1671999990940094,0.21305599808692932,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:15,0.6.3
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,2048,0.29440000653266907,0.2935679852962494,0.29523199796676636,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:15,0.6.3
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,3072,0.4164479970932007,0.4156160056591034,0.4176959991455078,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:15,0.6.3
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,4096,0.5370879769325256,0.5360320210456848,0.5380160212516785,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:15,0.6.3
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,5120,0.648576021194458,0.6475584149360657,0.6494399905204773,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:15,0.6.3
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,6144,0.7655680179595947,0.764415979385376,0.7665600180625916,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:15,0.6.3
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,7168,0.8751360177993774,0.8741440176963806,0.8761280179023743,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:15,0.6.3
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,8192,0.9919520020484924,0.990937626361847,0.9929919838905334,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:15,0.6.3
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,9216,1.0996160507202148,1.0985087871551515,1.100640058517456,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:15,0.6.3
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,10240,1.2153279781341553,1.214176058769226,1.2163200378417969,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:15,0.6.3
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,11264,1.3224480152130127,1.3213759660720825,1.323423981666565,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:15,0.6.3
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,12288,1.4376319646835327,1.4364800453186035,1.4387967824935912,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:15,0.6.3
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,13312,1.5448640584945679,1.5438208103179931,1.5462592124938965,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:15,0.6.3
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,14336,1.6600639820098877,1.6587712049484253,1.6611520051956177,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:15,0.6.3
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,15360,1.7677119970321655,1.7666688203811645,1.7688832044601441,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:15,0.6.3
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,16384,1.8829439878463745,1.8816319704055786,1.884108805656433,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:15,0.6.3
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,1024,0.16921600699424744,0.1666560024023056,0.20426879823207858,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:24,0.6.3
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,2048,0.294624000787735,0.2938239872455597,0.2953599989414215,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:24,0.6.3
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,3072,0.416159987449646,0.4153920114040375,0.41705599427223206,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:24,0.6.3
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,4096,0.5376960039138794,0.5366399884223938,0.5387840270996094,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:24,0.6.3
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,5120,0.6492159962654114,0.6483200192451477,0.650111973285675,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:24,0.6.3
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,6144,0.7663040161132812,0.7654399871826172,0.7670400142669678,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:24,0.6.3
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,7168,0.8759679794311523,0.8749120235443115,0.8768896222114563,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:24,0.6.3
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,8192,0.992576003074646,0.991487979888916,0.9936000108718872,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:24,0.6.3
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,9216,1.0997120141983032,1.0984575510025025,1.100640058517456,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:24,0.6.3
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,10240,1.215391993522644,1.2143679857254028,1.216275191307068,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:24,0.6.3
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,11264,1.322335958480835,1.3210879564285278,1.3234688282012939,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:24,0.6.3
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,12288,1.4377599954605103,1.4363199472427368,1.4388544082641603,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:24,0.6.3
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,13312,1.5448640584945679,1.5438400506973267,1.5457919836044312,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:24,0.6.3
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,14336,1.6605920195579529,1.6593536376953124,1.661798429489136,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:24,0.6.3
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,15360,1.7679040431976318,1.7666368007659912,1.7689152240753174,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:24,0.6.3
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,16384,1.882688045501709,1.881478452682495,1.88372483253479,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:24,0.6.3
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,1024,0.3615359961986542,0.33838080167770385,0.3977855920791626,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:32,0.6.3
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,2048,0.3851200044155121,0.3596160113811493,0.45038079619407656,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:32,0.6.3
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,3072,0.39024001359939575,0.36393600702285767,0.46221440434455874,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:32,0.6.3
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,4096,0.39049598574638367,0.363103985786438,0.46825599670410156,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:32,0.6.3
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,5120,0.38161601126194,0.35683199763298035,0.4332160115242004,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:32,0.6.3
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,6144,0.38576000928878784,0.3604223966598511,0.4454015910625458,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:32,0.6.3
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,7168,0.38447999954223633,0.36032000184059143,0.43699198961257935,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:32,0.6.3
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,8192,0.3773760050535202,0.35206401348114014,0.4291135907173157,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:32,0.6.3
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,9216,0.37700799107551575,0.35400319695472715,0.42578558921813964,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:32,0.6.3
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,10240,0.36979199945926666,0.35039359927177427,0.4096320033073425,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:32,0.6.3
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,11264,0.3845120072364807,0.35998080372810365,0.4341312050819397,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:32,0.6.3
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,12288,0.3848160058259964,0.36093440651893616,0.4542464077472687,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:32,0.6.3
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,13312,0.38705600798130035,0.3673535943031311,0.44297599792480474,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:32,0.6.3
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,14336,0.39051200449466705,0.365555202960968,0.4446080029010774,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:32,0.6.3
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,15360,0.39401599764823914,0.3699199855327606,0.45081600546836853,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:32,0.6.3
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,16384,0.3952319920063019,0.36908159255981443,0.4500671982765198,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:32,0.6.3
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,1024,0.2444000020623207,0.2260800004005432,0.30337279438972475,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:40,0.6.3
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,2048,0.4185279905796051,0.4177280068397522,0.41948801279067993,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:40,0.6.3
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,3072,0.5965120196342468,0.5955520272254944,0.5974720120429993,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:40,0.6.3
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,4096,0.7712960243225098,0.7700480222702026,0.7722880244255066,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:40,0.6.3
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,5120,0.9364479780197144,0.9355199933052063,0.9373120069503784,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:40,0.6.3
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,6144,1.10697603225708,1.1060800552368164,1.1079360246658325,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:40,0.6.3
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,7168,1.2694720029830933,1.2684032201766966,1.270527958869934,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:40,0.6.3
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,8192,1.4403200149536133,1.4393279552459717,1.441375970840454,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:40,0.6.3
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,9216,1.6009279489517212,1.5998272180557251,1.60207998752594,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:40,0.6.3
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,10240,1.7699520587921143,1.7683071851730348,1.770796799659729,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:40,0.6.3
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,11264,1.9306880235671997,1.9295040369033813,1.9319679737091064,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:40,0.6.3
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,12288,2.099423885345459,2.098035192489624,2.1005248069763187,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:40,0.6.3
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,13312,2.260175943374634,2.258905601501465,2.2613439559936523,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:40,0.6.3
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,14336,2.4294559955596924,2.4281599521636963,2.430630397796631,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:40,0.6.3
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,15360,2.5899839401245117,2.588729667663574,2.591091251373291,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:40,0.6.3
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,16384,2.7592639923095703,2.7580161094665527,2.760864019393921,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:40,0.6.3
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,1024,0.29423999786376953,0.27457280158996583,0.34389119148254393,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,2048,0.4188480079174042,0.41783680319786076,0.42273918986320497,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,3072,0.5990080237388611,0.5965631961822511,0.6032959818840027,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,4096,0.7752000093460083,0.7717440128326416,0.7791039943695068,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,5120,0.9398720264434814,0.9368000030517578,0.942636775970459,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,6144,1.1084799766540527,1.1065599918365479,1.1135295629501343,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,7168,1.2726080417633057,1.2696512222290037,1.276217555999756,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,8192,1.4449759721755981,1.4412863969802858,1.4484927892684936,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,9216,1.6035840511322021,1.6010368108749389,1.6065152168273926,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,10240,1.7722560167312622,1.7695295572280885,1.7771519422531128,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,11264,1.9326080083847046,1.9303104400634767,1.9362175941467286,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,12288,2.0995519161224365,2.098112106323242,2.100831985473633,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,13312,2.260319948196411,2.2591872215270996,2.2617088794708256,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,14336,2.4292320013046265,2.4280319213867188,2.430694341659546,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,15360,2.5894880294799805,2.58841609954834,2.5908351898193356,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,16384,2.7594720125198364,2.7581119537353516,2.7606592178344727,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,liger,full,memory,MB,hidden_size,hidden_size,1024,41.07177734375,41.07177734375,41.07177734375,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,liger,full,memory,MB,hidden_size,hidden_size,2048,82.14111328125,82.14111328125,82.14111328125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,liger,full,memory,MB,hidden_size,hidden_size,3072,123.21044921875,123.21044921875,123.21044921875,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,liger,full,memory,MB,hidden_size,hidden_size,4096,164.27978515625,164.27978515625,164.27978515625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,liger,full,memory,MB,hidden_size,hidden_size,5120,205.34912109375,205.34912109375,205.34912109375,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,liger,full,memory,MB,hidden_size,hidden_size,6144,246.41845703125,246.41845703125,246.41845703125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,liger,full,memory,MB,hidden_size,hidden_size,7168,287.48779296875,287.48779296875,287.48779296875,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,liger,full,memory,MB,hidden_size,hidden_size,8192,328.55712890625,328.55712890625,328.55712890625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,liger,full,memory,MB,hidden_size,hidden_size,9216,369.62646484375,369.62646484375,369.62646484375,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,liger,full,memory,MB,hidden_size,hidden_size,10240,410.69580078125,410.69580078125,410.69580078125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,liger,full,memory,MB,hidden_size,hidden_size,11264,451.76513671875,451.76513671875,451.76513671875,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,liger,full,memory,MB,hidden_size,hidden_size,12288,492.83447265625,492.83447265625,492.83447265625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,liger,full,memory,MB,hidden_size,hidden_size,13312,533.90380859375,533.90380859375,533.90380859375,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,liger,full,memory,MB,hidden_size,hidden_size,14336,575.75439453125,575.75439453125,575.75439453125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,liger,full,memory,MB,hidden_size,hidden_size,15360,616.30810546875,616.30810546875,616.30810546875,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,liger,full,memory,MB,hidden_size,hidden_size,16384,657.11181640625,657.11181640625,657.11181640625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch,full,memory,MB,hidden_size,hidden_size,1024,136.03759765625,136.03759765625,136.03759765625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch,full,memory,MB,hidden_size,hidden_size,2048,272.07275390625,272.07275390625,272.07275390625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch,full,memory,MB,hidden_size,hidden_size,3072,408.10791015625,408.10791015625,408.10791015625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch,full,memory,MB,hidden_size,hidden_size,4096,544.14306640625,544.14306640625,544.14306640625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch,full,memory,MB,hidden_size,hidden_size,5120,652.67822265625,652.67822265625,652.67822265625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch,full,memory,MB,hidden_size,hidden_size,6144,768.189453125,768.189453125,768.189453125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch,full,memory,MB,hidden_size,hidden_size,7168,896.220703125,896.220703125,896.220703125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch,full,memory,MB,hidden_size,hidden_size,8192,1024.251953125,1024.251953125,1024.251953125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch,full,memory,MB,hidden_size,hidden_size,9216,1152.283203125,1152.283203125,1152.283203125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch,full,memory,MB,hidden_size,hidden_size,10240,1280.314453125,1280.314453125,1280.314453125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch,full,memory,MB,hidden_size,hidden_size,11264,1408.345703125,1408.345703125,1408.345703125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch,full,memory,MB,hidden_size,hidden_size,12288,1536.376953125,1536.376953125,1536.376953125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch,full,memory,MB,hidden_size,hidden_size,13312,1664.408203125,1664.408203125,1664.408203125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch,full,memory,MB,hidden_size,hidden_size,14336,1792.439453125,1792.439453125,1792.439453125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch,full,memory,MB,hidden_size,hidden_size,15360,1920.470703125,1920.470703125,1920.470703125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch,full,memory,MB,hidden_size,hidden_size,16384,2048.501953125,2048.501953125,2048.501953125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,memory,MB,hidden_size,hidden_size,1024,136.03759765625,136.03759765625,136.03759765625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,memory,MB,hidden_size,hidden_size,2048,272.07275390625,272.07275390625,272.07275390625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,memory,MB,hidden_size,hidden_size,3072,408.10791015625,408.10791015625,408.10791015625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,memory,MB,hidden_size,hidden_size,4096,544.14306640625,544.14306640625,544.14306640625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,memory,MB,hidden_size,hidden_size,5120,652.67822265625,652.67822265625,652.67822265625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,memory,MB,hidden_size,hidden_size,6144,768.189453125,768.189453125,768.189453125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,memory,MB,hidden_size,hidden_size,7168,896.220703125,896.220703125,896.220703125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,memory,MB,hidden_size,hidden_size,8192,1024.251953125,1024.251953125,1024.251953125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,memory,MB,hidden_size,hidden_size,9216,1152.283203125,1152.283203125,1152.283203125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,memory,MB,hidden_size,hidden_size,10240,1280.314453125,1280.314453125,1280.314453125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,memory,MB,hidden_size,hidden_size,11264,1408.345703125,1408.345703125,1408.345703125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,memory,MB,hidden_size,hidden_size,12288,1536.376953125,1536.376953125,1536.376953125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,memory,MB,hidden_size,hidden_size,13312,1664.408203125,1664.408203125,1664.408203125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,memory,MB,hidden_size,hidden_size,14336,1792.439453125,1792.439453125,1792.439453125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,memory,MB,hidden_size,hidden_size,15360,1920.470703125,1920.470703125,1920.470703125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+dyt_beta=True,torch_compile,full,memory,MB,hidden_size,hidden_size,16384,2048.501953125,2048.501953125,2048.501953125,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-11-21 00:30:49,0.6.3
+embedding,liger,forward,speed,ms,V,embedding dimension,1024,0.02457600086927414,0.024320000782608986,0.025087999179959297,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:03,0.6.3
+embedding,liger,forward,speed,ms,V,embedding dimension,2048,0.025760000571608543,0.025536000728607178,0.02630399912595749,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:03,0.6.3
+embedding,liger,forward,speed,ms,V,embedding dimension,4096,0.02755199931561947,0.027295999228954315,0.028255999088287354,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:03,0.6.3
+embedding,liger,forward,speed,ms,V,embedding dimension,8192,0.030592000111937523,0.030271999537944794,0.031136000528931618,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:03,0.6.3
+embedding,liger,forward,speed,ms,V,embedding dimension,16384,0.0342399999499321,0.03386880084872246,0.034912001341581345,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:03,0.6.3
+embedding,liger,forward,speed,ms,V,embedding dimension,32768,0.037408001720905304,0.036896001547575,0.03810559958219528,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:03,0.6.3
+embedding,liger,forward,speed,ms,V,embedding dimension,65536,0.03884800150990486,0.03843199834227562,0.03949439898133278,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:03,0.6.3
+embedding,liger,forward,speed,ms,V,embedding dimension,131072,0.03993599861860275,0.039455998688936234,0.04064000025391579,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:03,0.6.3
+embedding,huggingface,forward,speed,ms,V,embedding dimension,1024,0.024064000695943832,0.023903999477624893,0.02454400062561035,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:06,0.6.3
+embedding,huggingface,forward,speed,ms,V,embedding dimension,2048,0.026688000187277794,0.026496000587940216,0.027103999629616737,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:06,0.6.3
+embedding,huggingface,forward,speed,ms,V,embedding dimension,4096,0.030368000268936157,0.030112000182271004,0.031072000041604042,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:06,0.6.3
+embedding,huggingface,forward,speed,ms,V,embedding dimension,8192,0.03494400158524513,0.034752000123262405,0.03574400022625923,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:06,0.6.3
+embedding,huggingface,forward,speed,ms,V,embedding dimension,16384,0.03807999938726425,0.03782400116324425,0.03884800150990486,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:06,0.6.3
+embedding,huggingface,forward,speed,ms,V,embedding dimension,32768,0.0395519994199276,0.03936000168323517,0.040352001786231995,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:06,0.6.3
+embedding,huggingface,forward,speed,ms,V,embedding dimension,65536,0.04047999903559685,0.04016000032424927,0.04124800115823746,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:06,0.6.3
+embedding,huggingface,forward,speed,ms,V,embedding dimension,131072,0.04095999896526337,0.04064000025391579,0.041760001331567764,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:06,0.6.3
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,1024,0.038176000118255615,0.035711999982595444,0.05127039924263954,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:15,0.6.3
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,2048,0.036639999598264694,0.035020798444747925,0.05079039931297302,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:15,0.6.3
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,4096,0.0379519984126091,0.03513599932193756,0.051513599604368215,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:15,0.6.3
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,8192,0.040608000010252,0.037171199917793274,0.04416640102863312,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:15,0.6.3
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,16384,0.04028800129890442,0.03856639936566353,0.05364479944109917,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:15,0.6.3
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,32768,0.04036800004541874,0.03929600119590759,0.04854400083422661,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:15,0.6.3
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,65536,0.04095999896526337,0.04016000032424927,0.044198401272296906,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:15,0.6.3
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,131072,0.041600000113248825,0.040832001715898514,0.04371200129389763,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:15,0.6.3
+embedding,liger,backward,speed,ms,V,embedding dimension,1024,0.14060800522565842,0.14009599685668944,0.14078080654144287,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:19,0.6.3
+embedding,liger,backward,speed,ms,V,embedding dimension,2048,0.14643199741840363,0.14498559534549713,0.14707200229167938,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:19,0.6.3
+embedding,liger,backward,speed,ms,V,embedding dimension,4096,0.15731200575828552,0.156031996011734,0.15799680054187776,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:19,0.6.3
+embedding,liger,backward,speed,ms,V,embedding dimension,8192,0.18063999712467194,0.1793919950723648,0.181536003947258,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:19,0.6.3
+embedding,liger,backward,speed,ms,V,embedding dimension,16384,0.22256000339984894,0.22127999365329742,0.2234880030155182,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:19,0.6.3
+embedding,liger,backward,speed,ms,V,embedding dimension,32768,0.29235199093818665,0.29120001196861267,0.29334399104118347,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:19,0.6.3
+embedding,liger,backward,speed,ms,V,embedding dimension,65536,0.42342400550842285,0.42214399576187134,0.42425599694252014,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:19,0.6.3
+embedding,liger,backward,speed,ms,V,embedding dimension,131072,0.6806399822235107,0.6793919801712036,0.6816959977149963,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:19,0.6.3
+embedding,huggingface,backward,speed,ms,V,embedding dimension,1024,0.20049600303173065,0.1995840072631836,0.20163199305534363,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:22,0.6.3
+embedding,huggingface,backward,speed,ms,V,embedding dimension,2048,0.2120320051908493,0.21127680242061614,0.21288320124149324,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:22,0.6.3
+embedding,huggingface,backward,speed,ms,V,embedding dimension,4096,0.24324800074100494,0.24246400594711304,0.2442624032497406,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:22,0.6.3
+embedding,huggingface,backward,speed,ms,V,embedding dimension,8192,0.3115839958190918,0.3108159899711609,0.312691193819046,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:22,0.6.3
+embedding,huggingface,backward,speed,ms,V,embedding dimension,16384,0.41467200219631195,0.4138688027858734,0.41563521027565004,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:22,0.6.3
+embedding,huggingface,backward,speed,ms,V,embedding dimension,32768,0.5027199983596802,0.5020480155944824,0.5039039850234985,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:22,0.6.3
+embedding,huggingface,backward,speed,ms,V,embedding dimension,65536,0.6472800076007843,0.6464639902114868,0.6483200192451477,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:22,0.6.3
+embedding,huggingface,backward,speed,ms,V,embedding dimension,131072,0.9112800061702728,0.9103360176086426,0.912992000579834,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:22,0.6.3
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,1024,0.17334400117397308,0.15360000729560852,0.21536639630794524,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:30,0.6.3
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,2048,0.16259199380874634,0.14873600006103516,0.18649600446224213,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:30,0.6.3
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,4096,0.16150400042533875,0.14942079782485962,0.20946560502052308,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:30,0.6.3
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,8192,0.16732800006866455,0.14905600249767303,0.22051199972629548,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:30,0.6.3
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,16384,0.17335999757051468,0.15847040414810182,0.20698239803314208,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:30,0.6.3
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,32768,0.2260800004005432,0.22462080419063568,0.22706559896469114,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:30,0.6.3
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,65536,0.3563840091228485,0.354886394739151,0.35737600922584534,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:30,0.6.3
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,131072,0.6148160099983215,0.613433587551117,0.6157183885574341,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:30,0.6.3
+embedding,liger,full,speed,ms,V,embedding dimension,1024,0.22884800285100937,0.20997120440006256,0.3053695857524872,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:33,0.6.3
+embedding,liger,full,speed,ms,V,embedding dimension,2048,0.22460800409317017,0.2011391967535019,0.280185604095459,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:33,0.6.3
+embedding,liger,full,speed,ms,V,embedding dimension,4096,0.21244800090789795,0.19628800451755524,0.2758080065250397,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:33,0.6.3
+embedding,liger,full,speed,ms,V,embedding dimension,8192,0.21052800118923187,0.2092544049024582,0.23766399621963502,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:33,0.6.3
+embedding,liger,full,speed,ms,V,embedding dimension,16384,0.2543039917945862,0.2531520128250122,0.2554816067218781,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:33,0.6.3
+embedding,liger,full,speed,ms,V,embedding dimension,32768,0.327455997467041,0.3266304075717926,0.3283776104450226,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:33,0.6.3
+embedding,liger,full,speed,ms,V,embedding dimension,65536,0.45983999967575073,0.4588479995727539,0.4610559940338135,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:33,0.6.3
+embedding,liger,full,speed,ms,V,embedding dimension,131072,0.7188159823417664,0.7179840207099915,0.7198655962944032,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:33,0.6.3
+embedding,huggingface,full,speed,ms,V,embedding dimension,1024,0.22353599965572357,0.22188800573349,0.25913599133491516,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:36,0.6.3
+embedding,huggingface,full,speed,ms,V,embedding dimension,2048,0.23747199773788452,0.2367040067911148,0.2385919988155365,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:36,0.6.3
+embedding,huggingface,full,speed,ms,V,embedding dimension,4096,0.272352010011673,0.27139198780059814,0.2733759880065918,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:36,0.6.3
+embedding,huggingface,full,speed,ms,V,embedding dimension,8192,0.344543993473053,0.3436543881893158,0.3455680012702942,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:36,0.6.3
+embedding,huggingface,full,speed,ms,V,embedding dimension,16384,0.4505920112133026,0.44972801208496094,0.45177599787712097,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:36,0.6.3
+embedding,huggingface,full,speed,ms,V,embedding dimension,32768,0.5408799946308136,0.5397120118141174,0.5420991897583007,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:36,0.6.3
+embedding,huggingface,full,speed,ms,V,embedding dimension,65536,0.6856319904327393,0.6844480037689209,0.6868608236312865,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:36,0.6.3
+embedding,huggingface,full,speed,ms,V,embedding dimension,131072,0.9491519927978516,0.9481023907661438,0.9507840275764465,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:36,0.6.3
+embedding,torch_compile,full,speed,ms,V,embedding dimension,1024,0.3606080114841461,0.337388801574707,0.401420795917511,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:39,0.6.3
+embedding,torch_compile,full,speed,ms,V,embedding dimension,2048,0.3620480000972748,0.3361983895301819,0.43664001226425175,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:39,0.6.3
+embedding,torch_compile,full,speed,ms,V,embedding dimension,4096,0.358271986246109,0.3325119912624359,0.41501439809799195,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:39,0.6.3
+embedding,torch_compile,full,speed,ms,V,embedding dimension,8192,0.37225601077079773,0.3444800078868866,0.4283519983291626,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:39,0.6.3
+embedding,torch_compile,full,speed,ms,V,embedding dimension,16384,0.36110401153564453,0.3370751917362213,0.41931519508361814,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:39,0.6.3
+embedding,torch_compile,full,speed,ms,V,embedding dimension,32768,0.3894079923629761,0.3565056025981903,0.534060788154602,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:39,0.6.3
+embedding,torch_compile,full,speed,ms,V,embedding dimension,65536,0.3946239948272705,0.39306240677833554,0.39832319021224977,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:39,0.6.3
+embedding,torch_compile,full,speed,ms,V,embedding dimension,131072,0.6520480215549469,0.6510016083717346,0.653382384777069,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:39,0.6.3
+embedding,liger,forward,speed,ms,V,embedding dimension,1024,0.10598400235176086,0.10556799918413162,0.10649599879980087,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:52,0.6.3
+embedding,liger,forward,speed,ms,V,embedding dimension,2048,0.11356800049543381,0.11321599781513214,0.11398400366306305,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:52,0.6.3
+embedding,liger,forward,speed,ms,V,embedding dimension,4096,0.12486399710178375,0.12441600114107132,0.12556800246238708,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:52,0.6.3
+embedding,liger,forward,speed,ms,V,embedding dimension,8192,0.14243200421333313,0.14188799262046814,0.14310400187969208,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:52,0.6.3
+embedding,liger,forward,speed,ms,V,embedding dimension,16384,0.16096000373363495,0.16016000509262085,0.1616320013999939,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:52,0.6.3
+embedding,liger,forward,speed,ms,V,embedding dimension,32768,0.17395199835300446,0.17318400740623474,0.17443199455738068,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:52,0.6.3
+embedding,liger,forward,speed,ms,V,embedding dimension,65536,0.18352000415325165,0.18294399976730347,0.1839359998703003,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:52,0.6.3
+embedding,liger,forward,speed,ms,V,embedding dimension,131072,0.18777599930763245,0.1872575968503952,0.18822400271892548,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:31:52,0.6.3
+embedding,huggingface,forward,speed,ms,V,embedding dimension,1024,0.10812799632549286,0.10787200182676315,0.10844799876213074,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:05,0.6.3
+embedding,huggingface,forward,speed,ms,V,embedding dimension,2048,0.12377600371837616,0.12355200201272964,0.12425599992275238,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:05,0.6.3
+embedding,huggingface,forward,speed,ms,V,embedding dimension,4096,0.1456640064716339,0.14524799585342407,0.14633600413799286,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:05,0.6.3
+embedding,huggingface,forward,speed,ms,V,embedding dimension,8192,0.1656319946050644,0.16489599645137787,0.16604800522327423,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:05,0.6.3
+embedding,huggingface,forward,speed,ms,V,embedding dimension,16384,0.17830400168895721,0.17766399681568146,0.17865599691867828,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:05,0.6.3
+embedding,huggingface,forward,speed,ms,V,embedding dimension,32768,0.18454399704933167,0.18403199315071106,0.18486399948596954,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:05,0.6.3
+embedding,huggingface,forward,speed,ms,V,embedding dimension,65536,0.18860800564289093,0.18818559646606445,0.18883199989795685,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:05,0.6.3
+embedding,huggingface,forward,speed,ms,V,embedding dimension,131072,0.19036799669265747,0.18998399376869202,0.19065600633621216,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:05,0.6.3
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,1024,0.10787200182676315,0.10758399963378906,0.10819199681282043,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:17,0.6.3
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,2048,0.12412799894809723,0.12383999675512314,0.12454400211572647,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:17,0.6.3
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,4096,0.14553600549697876,0.1451520025730133,0.14634880125522615,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:17,0.6.3
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,8192,0.16579200327396393,0.1650560051202774,0.16630400717258453,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:17,0.6.3
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,16384,0.17750400304794312,0.17679359912872314,0.177824005484581,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:17,0.6.3
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,32768,0.18486399948596954,0.18432000279426575,0.18512000143527985,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:17,0.6.3
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,65536,0.1879040002822876,0.18751999735832214,0.18822400271892548,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:17,0.6.3
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,131072,0.19033600389957428,0.18988800048828125,0.19068799912929535,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:17,0.6.3
+embedding,liger,backward,speed,ms,V,embedding dimension,1024,0.6760639846324921,0.6745728135108948,0.6768831968307495,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:30,0.6.3
+embedding,liger,backward,speed,ms,V,embedding dimension,2048,0.7120640277862549,0.7107840180397034,0.7130879759788513,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:30,0.6.3
+embedding,liger,backward,speed,ms,V,embedding dimension,4096,0.7807679772377014,0.7792575836181641,0.7818880081176758,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:30,0.6.3
+embedding,liger,backward,speed,ms,V,embedding dimension,8192,0.8960000276565552,0.894815981388092,0.8975039720535278,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:30,0.6.3
+embedding,liger,backward,speed,ms,V,embedding dimension,16384,1.093616008758545,1.0920384168624877,1.0945663928985594,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:30,0.6.3
+embedding,liger,backward,speed,ms,V,embedding dimension,32768,1.4537919759750366,1.4520831823348999,1.4551040410995484,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:30,0.6.3
+embedding,liger,backward,speed,ms,V,embedding dimension,65536,2.1452159881591797,2.1443968296051024,2.146892786026001,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:30,0.6.3
+embedding,liger,backward,speed,ms,V,embedding dimension,131072,3.5155200958251953,3.51379189491272,3.5166271686553956,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:30,0.6.3
+embedding,huggingface,backward,speed,ms,V,embedding dimension,1024,0.4739840030670166,0.4732159972190857,0.4750719964504242,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:43,0.6.3
+embedding,huggingface,backward,speed,ms,V,embedding dimension,2048,0.5377600193023682,0.5369279980659485,0.5387584209442139,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:43,0.6.3
+embedding,huggingface,backward,speed,ms,V,embedding dimension,4096,0.6866079866886139,0.6854207992553711,0.6876864075660706,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:43,0.6.3
+embedding,huggingface,backward,speed,ms,V,embedding dimension,8192,0.9743679761886597,0.9731136083602905,0.9757632136344909,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:43,0.6.3
+embedding,huggingface,backward,speed,ms,V,embedding dimension,16384,1.414192020893097,1.4129600524902344,1.4158079624176025,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:43,0.6.3
+embedding,huggingface,backward,speed,ms,V,embedding dimension,32768,1.8453760147094727,1.8440639972686768,1.8468799591064453,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:43,0.6.3
+embedding,huggingface,backward,speed,ms,V,embedding dimension,65536,2.580415964126587,2.578847885131836,2.581971216201782,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:43,0.6.3
+embedding,huggingface,backward,speed,ms,V,embedding dimension,131072,3.975856065750122,3.974835157394409,3.978867244720459,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:43,0.6.3
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,1024,0.4747679978609085,0.4739519953727722,0.4761599898338318,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:55,0.6.3
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,2048,0.5382400155067444,0.5372991800308228,0.5392384171485901,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:55,0.6.3
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,4096,0.686784029006958,0.6857600212097168,0.6887296199798585,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:55,0.6.3
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,8192,0.9700160026550293,0.9687936067581177,0.9713919758796692,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:55,0.6.3
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,16384,1.416159987449646,1.415071964263916,1.4175360202789307,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:55,0.6.3
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,32768,1.844864010810852,1.8435839414596558,1.8464640378952026,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:55,0.6.3
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,65536,2.581439971923828,2.580044889450073,2.584723138809204,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:55,0.6.3
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,131072,3.9759199619293213,3.9738112449645997,3.9780033111572264,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:32:55,0.6.3
+embedding,liger,full,speed,ms,V,embedding dimension,1024,0.7800319790840149,0.7790719866752625,0.7809152007102966,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:08,0.6.3
+embedding,liger,full,speed,ms,V,embedding dimension,2048,0.8235200047492981,0.8223424196243285,0.8245568156242371,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:08,0.6.3
+embedding,liger,full,speed,ms,V,embedding dimension,4096,0.9024959802627563,0.9014719724655151,0.9037439823150635,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:08,0.6.3
+embedding,liger,full,speed,ms,V,embedding dimension,8192,1.0361599922180176,1.0352959632873535,1.0379136085510254,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:08,0.6.3
+embedding,liger,full,speed,ms,V,embedding dimension,16384,1.2517600059509277,1.250611162185669,1.252992033958435,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:08,0.6.3
+embedding,liger,full,speed,ms,V,embedding dimension,32768,1.6260160207748413,1.6246527671813964,1.6267967700958252,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:08,0.6.3
+embedding,liger,full,speed,ms,V,embedding dimension,65536,2.3247039318084717,2.323904037475586,2.326240062713623,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:08,0.6.3
+embedding,liger,full,speed,ms,V,embedding dimension,131072,3.7003999948501587,3.6993279457092285,3.7028799057006836,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:08,0.6.3
+embedding,huggingface,full,speed,ms,V,embedding dimension,1024,0.5806400179862976,0.5799935817718506,0.5816704154014587,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:22,0.6.3
+embedding,huggingface,full,speed,ms,V,embedding dimension,2048,0.6602879762649536,0.6592640280723572,0.6615999937057495,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:22,0.6.3
+embedding,huggingface,full,speed,ms,V,embedding dimension,4096,0.8316479921340942,0.8300607800483704,1.293862390518189,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:22,0.6.3
+embedding,huggingface,full,speed,ms,V,embedding dimension,8192,1.1360000371932983,1.1352640390396118,1.1376960277557373,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:22,0.6.3
+embedding,huggingface,full,speed,ms,V,embedding dimension,16384,1.5893440246582031,1.5873279809951781,1.5908159732818605,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:22,0.6.3
+embedding,huggingface,full,speed,ms,V,embedding dimension,32768,2.0288960933685303,2.0271168231964114,2.0318528175354005,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:22,0.6.3
+embedding,huggingface,full,speed,ms,V,embedding dimension,65536,2.766752004623413,2.765004825592041,2.7680895805358885,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:22,0.6.3
+embedding,huggingface,full,speed,ms,V,embedding dimension,131072,4.167520046234131,4.164307022094726,4.169990539550781,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:22,0.6.3
+embedding,torch_compile,full,speed,ms,V,embedding dimension,1024,0.5805599987506866,0.5797119736671448,0.581279993057251,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:34,0.6.3
+embedding,torch_compile,full,speed,ms,V,embedding dimension,2048,0.6592640280723572,0.6583295822143554,0.6605119705200195,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:34,0.6.3
+embedding,torch_compile,full,speed,ms,V,embedding dimension,4096,0.8309279978275299,0.8300544023513794,0.8323007941246032,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:34,0.6.3
+embedding,torch_compile,full,speed,ms,V,embedding dimension,8192,1.1341439485549927,1.133075213432312,1.1357824087142945,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:34,0.6.3
+embedding,torch_compile,full,speed,ms,V,embedding dimension,16384,1.5911680459976196,1.5891007900238037,1.5924927711486816,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:34,0.6.3
+embedding,torch_compile,full,speed,ms,V,embedding dimension,32768,2.028736114501953,2.027782440185547,2.02988805770874,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:34,0.6.3
+embedding,torch_compile,full,speed,ms,V,embedding dimension,65536,2.7669761180877686,2.76627197265625,2.7683968544006348,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:34,0.6.3
+embedding,torch_compile,full,speed,ms,V,embedding dimension,131072,4.164383888244629,4.160927867889404,4.1695297241210945,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:34,0.6.3
+embedding,liger,full,memory,MB,V,embedding dimension,1024,108.125,108.125,108.125,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:36,0.6.3
+embedding,liger,full,memory,MB,V,embedding dimension,2048,120.125,120.125,120.125,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:36,0.6.3
+embedding,liger,full,memory,MB,V,embedding dimension,4096,144.125,144.125,144.125,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:36,0.6.3
+embedding,liger,full,memory,MB,V,embedding dimension,8192,192.125,192.125,192.125,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:36,0.6.3
+embedding,liger,full,memory,MB,V,embedding dimension,16384,288.125,288.125,288.125,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:36,0.6.3
+embedding,liger,full,memory,MB,V,embedding dimension,32768,480.125,480.125,480.125,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:36,0.6.3
+embedding,liger,full,memory,MB,V,embedding dimension,65536,864.125,864.125,864.125,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:36,0.6.3
+embedding,liger,full,memory,MB,V,embedding dimension,131072,1632.125,1632.125,1632.125,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:36,0.6.3
+embedding,huggingface,full,memory,MB,V,embedding dimension,1024,116.537109375,116.537109375,116.537109375,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:39,0.6.3
+embedding,huggingface,full,memory,MB,V,embedding dimension,2048,131.359375,131.359375,131.359375,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:39,0.6.3
+embedding,huggingface,full,memory,MB,V,embedding dimension,4096,161.40625,161.40625,161.40625,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:39,0.6.3
+embedding,huggingface,full,memory,MB,V,embedding dimension,8192,221.5,221.5,221.5,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:39,0.6.3
+embedding,huggingface,full,memory,MB,V,embedding dimension,16384,341.6875,341.6875,341.6875,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:39,0.6.3
+embedding,huggingface,full,memory,MB,V,embedding dimension,32768,533.6875,533.6875,533.6875,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:39,0.6.3
+embedding,huggingface,full,memory,MB,V,embedding dimension,65536,917.6875,917.6875,917.6875,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:39,0.6.3
+embedding,huggingface,full,memory,MB,V,embedding dimension,131072,1685.6875,1685.6875,1685.6875,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:39,0.6.3
+embedding,torch_compile,full,memory,MB,V,embedding dimension,1024,108.125,108.125,108.125,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:41,0.6.3
+embedding,torch_compile,full,memory,MB,V,embedding dimension,2048,120.125,120.125,120.125,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:41,0.6.3
+embedding,torch_compile,full,memory,MB,V,embedding dimension,4096,144.125,144.125,144.125,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:41,0.6.3
+embedding,torch_compile,full,memory,MB,V,embedding dimension,8192,192.125,192.125,192.125,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:41,0.6.3
+embedding,torch_compile,full,memory,MB,V,embedding dimension,16384,288.125,288.125,288.125,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:41,0.6.3
+embedding,torch_compile,full,memory,MB,V,embedding dimension,32768,480.125,480.125,480.125,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:41,0.6.3
+embedding,torch_compile,full,memory,MB,V,embedding dimension,65536,864.125,864.125,864.125,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:41,0.6.3
+embedding,torch_compile,full,memory,MB,V,embedding dimension,131072,1632.125,1632.125,1632.125,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:41,0.6.3
+embedding,liger,full,memory,MB,V,embedding dimension,1024,576.125,576.125,576.125,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:52,0.6.3
+embedding,liger,full,memory,MB,V,embedding dimension,2048,640.125,640.125,640.125,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:52,0.6.3
+embedding,liger,full,memory,MB,V,embedding dimension,4096,768.125,768.125,768.125,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:52,0.6.3
+embedding,liger,full,memory,MB,V,embedding dimension,8192,1024.125,1024.125,1024.125,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:52,0.6.3
+embedding,liger,full,memory,MB,V,embedding dimension,16384,1536.125,1536.125,1536.125,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:52,0.6.3
+embedding,liger,full,memory,MB,V,embedding dimension,32768,2560.125,2560.125,2560.125,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:52,0.6.3
+embedding,liger,full,memory,MB,V,embedding dimension,65536,4608.125,4608.125,4608.125,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:52,0.6.3
+embedding,liger,full,memory,MB,V,embedding dimension,131072,8704.125,8704.125,8704.125,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:33:52,0.6.3
+embedding,huggingface,full,memory,MB,V,embedding dimension,1024,618.130859375,618.130859375,618.130859375,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:04,0.6.3
+embedding,huggingface,full,memory,MB,V,embedding dimension,2048,698.154296875,698.154296875,698.154296875,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:04,0.6.3
+embedding,huggingface,full,memory,MB,V,embedding dimension,4096,858.201171875,858.201171875,858.201171875,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:04,0.6.3
+embedding,huggingface,full,memory,MB,V,embedding dimension,8192,1178.294921875,1178.294921875,1178.294921875,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:04,0.6.3
+embedding,huggingface,full,memory,MB,V,embedding dimension,16384,1818.482421875,1818.482421875,1818.482421875,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:04,0.6.3
+embedding,huggingface,full,memory,MB,V,embedding dimension,32768,2842.482421875,2842.482421875,2842.482421875,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:04,0.6.3
+embedding,huggingface,full,memory,MB,V,embedding dimension,65536,4890.482421875,4890.482421875,4890.482421875,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:04,0.6.3
+embedding,huggingface,full,memory,MB,V,embedding dimension,131072,8986.482421875,8986.482421875,8986.482421875,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:04,0.6.3
+embedding,torch_compile,full,memory,MB,V,embedding dimension,1024,618.130859375,618.130859375,618.130859375,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:16,0.6.3
+embedding,torch_compile,full,memory,MB,V,embedding dimension,2048,698.154296875,698.154296875,698.154296875,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:16,0.6.3
+embedding,torch_compile,full,memory,MB,V,embedding dimension,4096,858.201171875,858.201171875,858.201171875,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:16,0.6.3
+embedding,torch_compile,full,memory,MB,V,embedding dimension,8192,1178.294921875,1178.294921875,1178.294921875,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:16,0.6.3
+embedding,torch_compile,full,memory,MB,V,embedding dimension,16384,1818.482421875,1818.482421875,1818.482421875,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:16,0.6.3
+embedding,torch_compile,full,memory,MB,V,embedding dimension,32768,2842.482421875,2842.482421875,2842.482421875,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:16,0.6.3
+embedding,torch_compile,full,memory,MB,V,embedding dimension,65536,4890.482421875,4890.482421875,4890.482421875,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:16,0.6.3
+embedding,torch_compile,full,memory,MB,V,embedding dimension,131072,8986.482421875,8986.482421875,8986.482421875,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:16,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,forward,speed,ms,H,hidden size,1024,0.02675200067460537,0.02319999970495701,0.04111360087990762,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:25,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,forward,speed,ms,H,hidden size,2048,0.02937600016593933,0.02848000079393387,0.034752000123262405,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:25,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,forward,speed,ms,H,hidden size,4096,0.05049600079655647,0.0498879998922348,0.051231998950242996,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:25,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,forward,speed,ms,H,hidden size,8192,0.09536000341176987,0.09504000097513199,0.09558399766683578,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:25,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,forward,speed,ms,H,hidden size,16384,0.18409599363803864,0.1833920031785965,0.18451200425624847,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:25,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,forward,speed,ms,H,hidden size,32768,0.4343999922275543,0.43372800946235657,0.435263991355896,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:25,0.6.3
+fused_add_rms_norm,huggingface,forward,speed,ms,H,hidden size,1024,0.04896000027656555,0.048608001321554184,0.04947200044989586,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:28,0.6.3
+fused_add_rms_norm,huggingface,forward,speed,ms,H,hidden size,2048,0.07727999985218048,0.07651200145483017,0.07804799824953079,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:28,0.6.3
+fused_add_rms_norm,huggingface,forward,speed,ms,H,hidden size,4096,0.15305599570274353,0.15228800475597382,0.15379199385643005,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:28,0.6.3
+fused_add_rms_norm,huggingface,forward,speed,ms,H,hidden size,8192,0.2744959890842438,0.27379199862480164,0.2752319872379303,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:28,0.6.3
+fused_add_rms_norm,huggingface,forward,speed,ms,H,hidden size,16384,0.5130559802055359,0.5123519897460938,0.5139200091362,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:28,0.6.3
+fused_add_rms_norm,huggingface,forward,speed,ms,H,hidden size,32768,0.9862240254878998,0.9854080080986023,0.98716801404953,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:28,0.6.3
+fused_add_rms_norm,liger_rms_norm,forward,speed,ms,H,hidden size,1024,0.04604800045490265,0.0427200011909008,0.060876801609992995,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:32,0.6.3
+fused_add_rms_norm,liger_rms_norm,forward,speed,ms,H,hidden size,2048,0.04867200180888176,0.04732799902558327,0.06035200133919716,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:32,0.6.3
+fused_add_rms_norm,liger_rms_norm,forward,speed,ms,H,hidden size,4096,0.09516800194978714,0.09465599805116653,0.09567999839782715,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:32,0.6.3
+fused_add_rms_norm,liger_rms_norm,forward,speed,ms,H,hidden size,8192,0.17849600315093994,0.1777919977903366,0.17894400656223297,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:32,0.6.3
+fused_add_rms_norm,liger_rms_norm,forward,speed,ms,H,hidden size,16384,0.34431999921798706,0.34377598762512207,0.34515199065208435,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:32,0.6.3
+fused_add_rms_norm,liger_rms_norm,forward,speed,ms,H,hidden size,32768,0.68476802110672,0.6840000152587891,0.6854720115661621,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:32,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,full,speed,ms,H,hidden size,1024,0.3498240113258362,0.3271872103214264,0.38464000821113586,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:37,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,full,speed,ms,H,hidden size,2048,0.348224014043808,0.32515838742256165,0.39096319675445557,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:37,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,full,speed,ms,H,hidden size,4096,0.3476639986038208,0.3244992017745972,0.38140159845352173,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:37,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,full,speed,ms,H,hidden size,8192,0.3627679944038391,0.3429120063781739,0.40875520110130314,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:37,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,full,speed,ms,H,hidden size,16384,1.270031988620758,1.2683392047882078,1.2720320224761963,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:37,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,full,speed,ms,H,hidden size,32768,1.9323359727859497,1.9285823822021484,1.9372223854064943,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:37,0.6.3
+fused_add_rms_norm,huggingface,full,speed,ms,H,hidden size,1024,0.33616000413894653,0.31754239797592165,0.3699072003364563,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:40,0.6.3
+fused_add_rms_norm,huggingface,full,speed,ms,H,hidden size,2048,0.3372480124235153,0.31665919423103334,0.3819519937038422,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:40,0.6.3
+fused_add_rms_norm,huggingface,full,speed,ms,H,hidden size,4096,0.5503680109977722,0.5483520030975342,0.5526464223861695,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:40,0.6.3
+fused_add_rms_norm,huggingface,full,speed,ms,H,hidden size,8192,0.9999359846115112,0.9978560209274292,1.0024447917938233,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:40,0.6.3
+fused_add_rms_norm,huggingface,full,speed,ms,H,hidden size,16384,1.8853440284729004,1.8832703590393067,1.8877375602722168,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:40,0.6.3
+fused_add_rms_norm,huggingface,full,speed,ms,H,hidden size,32768,3.6456639766693115,3.64271993637085,3.647552013397217,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:40,0.6.3
+fused_add_rms_norm,liger_rms_norm,full,speed,ms,H,hidden size,1024,0.46348801255226135,0.43592320084571834,0.5044863820075989,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:44,0.6.3
+fused_add_rms_norm,liger_rms_norm,full,speed,ms,H,hidden size,2048,0.46592000126838684,0.43825280070304873,0.502892816066742,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:44,0.6.3
+fused_add_rms_norm,liger_rms_norm,full,speed,ms,H,hidden size,4096,0.47040000557899475,0.4458880126476288,0.5127999782562256,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:44,0.6.3
+fused_add_rms_norm,liger_rms_norm,full,speed,ms,H,hidden size,8192,0.565887987613678,0.5650240182876587,0.5670719742774963,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:44,0.6.3
+fused_add_rms_norm,liger_rms_norm,full,speed,ms,H,hidden size,16384,1.1107040047645569,1.1093759536743164,1.1122816085815428,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:44,0.6.3
+fused_add_rms_norm,liger_rms_norm,full,speed,ms,H,hidden size,32768,2.728640079498291,2.724735975265503,2.7326016426086426,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:44,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,backward,speed,ms,H,hidden size,1024,0.1921280026435852,0.17478400468826294,0.24124160408973694,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:47,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,backward,speed,ms,H,hidden size,2048,0.18502399325370789,0.17112960517406464,0.2246912032365799,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:47,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,backward,speed,ms,H,hidden size,4096,0.18995200097560883,0.17378559708595276,0.2313279956579209,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:47,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,backward,speed,ms,H,hidden size,8192,0.1960960030555725,0.17980800271034242,0.23805439770221712,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:47,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,backward,speed,ms,H,hidden size,16384,1.0877759456634521,1.0859456300735473,1.0896639823913574,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:47,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,backward,speed,ms,H,hidden size,32768,1.5058879852294922,1.5022528171539307,1.5102784156799316,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:47,0.6.3
+fused_add_rms_norm,huggingface,backward,speed,ms,H,hidden size,1024,0.21884800493717194,0.20403839647769928,0.2575936019420624,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:50,0.6.3
+fused_add_rms_norm,huggingface,backward,speed,ms,H,hidden size,2048,0.21478399634361267,0.21251200139522552,0.22537599503993988,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:50,0.6.3
+fused_add_rms_norm,huggingface,backward,speed,ms,H,hidden size,4096,0.398144006729126,0.3969919979572296,0.3993600010871887,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:50,0.6.3
+fused_add_rms_norm,huggingface,backward,speed,ms,H,hidden size,8192,0.7269120216369629,0.725600004196167,0.72816002368927,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:50,0.6.3
+fused_add_rms_norm,huggingface,backward,speed,ms,H,hidden size,16384,1.3735040426254272,1.3721535682678223,1.3750720024108887,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:50,0.6.3
+fused_add_rms_norm,huggingface,backward,speed,ms,H,hidden size,32768,2.66048002243042,2.6590399742126465,2.6622464179992673,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:50,0.6.3
+fused_add_rms_norm,liger_rms_norm,backward,speed,ms,H,hidden size,1024,0.26078400015830994,0.23968640565872193,0.3131712079048157,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_add_rms_norm,liger_rms_norm,backward,speed,ms,H,hidden size,2048,0.26310399174690247,0.24107519984245301,0.31884160041809084,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_add_rms_norm,liger_rms_norm,backward,speed,ms,H,hidden size,4096,0.25252801179885864,0.23680000007152557,0.2890367925167084,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_add_rms_norm,liger_rms_norm,backward,speed,ms,H,hidden size,8192,0.38979199528694153,0.3883199989795685,0.3912000060081482,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_add_rms_norm,liger_rms_norm,backward,speed,ms,H,hidden size,16384,0.7716479897499084,0.7699648022651673,0.7735551834106444,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_add_rms_norm,liger_rms_norm,backward,speed,ms,H,hidden size,32768,2.0471999645233154,2.0435519218444824,2.0514559745788574,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,full,memory,MB,H,hidden size,1024,72.546875,72.546875,72.546875,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,full,memory,MB,H,hidden size,2048,145.0859375,145.0859375,145.0859375,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,full,memory,MB,H,hidden size,4096,290.1640625,290.1640625,290.1640625,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,full,memory,MB,H,hidden size,8192,580.3203125,580.3203125,580.3203125,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,full,memory,MB,H,hidden size,16384,1160.6328125,1160.6328125,1160.6328125,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_add_rms_norm,liger_fused_add_rms_norm,full,memory,MB,H,hidden size,32768,2321.2578125,2321.2578125,2321.2578125,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_add_rms_norm,huggingface,full,memory,MB,H,hidden size,1024,104.03173828125,104.03173828125,104.03173828125,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_add_rms_norm,huggingface,full,memory,MB,H,hidden size,2048,208.05517578125,208.05517578125,208.05517578125,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_add_rms_norm,huggingface,full,memory,MB,H,hidden size,4096,416.10205078125,416.10205078125,416.10205078125,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_add_rms_norm,huggingface,full,memory,MB,H,hidden size,8192,832.19580078125,832.19580078125,832.19580078125,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_add_rms_norm,huggingface,full,memory,MB,H,hidden size,16384,1664.3125,1664.3125,1664.3125,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_add_rms_norm,huggingface,full,memory,MB,H,hidden size,32768,3328.625,3328.625,3328.625,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_add_rms_norm,liger_rms_norm,full,memory,MB,H,hidden size,1024,104.03564453125,104.03564453125,104.03564453125,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_add_rms_norm,liger_rms_norm,full,memory,MB,H,hidden size,2048,208.06298828125,208.06298828125,208.06298828125,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_add_rms_norm,liger_rms_norm,full,memory,MB,H,hidden size,4096,416.11767578125,416.11767578125,416.11767578125,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_add_rms_norm,liger_rms_norm,full,memory,MB,H,hidden size,8192,832.22705078125,832.22705078125,832.22705078125,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_add_rms_norm,liger_rms_norm,full,memory,MB,H,hidden size,16384,1544.44580078125,1544.44580078125,1544.44580078125,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_add_rms_norm,liger_rms_norm,full,memory,MB,H,hidden size,32768,2960.8837890625,2960.8837890625,2960.8837890625,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:34:53,0.6.3
+fused_linear_cross_entropy,liger,forward,speed,ms,BT,B x T,4096,155.091552734375,155.091552734375,155.091552734375,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:35:17,0.6.3
+fused_linear_cross_entropy,liger,forward,speed,ms,BT,B x T,8192,160.18006896972656,160.18006896972656,160.18006896972656,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:35:17,0.6.3
+fused_linear_cross_entropy,liger,forward,speed,ms,BT,B x T,16384,191.88265991210938,191.88265991210938,191.88265991210938,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:35:17,0.6.3
+fused_linear_cross_entropy,liger,forward,speed,ms,BT,B x T,32768,270.42376708984375,270.42376708984375,270.42376708984375,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:35:17,0.6.3
+fused_linear_cross_entropy,liger-fp32-accum,forward,speed,ms,BT,B x T,4096,156.38809204101562,156.38809204101562,156.38809204101562,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:35:35,0.6.3
+fused_linear_cross_entropy,liger-fp32-accum,forward,speed,ms,BT,B x T,8192,160.86019897460938,160.86019897460938,160.86019897460938,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:35:35,0.6.3
+fused_linear_cross_entropy,liger-fp32-accum,forward,speed,ms,BT,B x T,16384,192.67027282714844,192.67027282714844,192.67027282714844,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:35:35,0.6.3
+fused_linear_cross_entropy,liger-fp32-accum,forward,speed,ms,BT,B x T,32768,264.24212646484375,264.24212646484375,264.24212646484375,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:35:35,0.6.3
+fused_linear_cross_entropy,huggingface,forward,speed,ms,BT,B x T,4096,7.0486719608306885,6.973657798767089,7.049792003631592,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:35:48,0.6.3
+fused_linear_cross_entropy,huggingface,forward,speed,ms,BT,B x T,8192,14.10921573638916,13.960479927062988,14.552908897399902,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:35:48,0.6.3
+fused_linear_cross_entropy,huggingface,forward,speed,ms,BT,B x T,16384,27.695423126220703,27.54188117980957,27.924172210693357,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:35:48,0.6.3
+fused_linear_cross_entropy,huggingface,forward,speed,ms,BT,B x T,32768,56.268672943115234,56.268672943115234,56.268672943115234,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:35:48,0.6.3
+fused_linear_cross_entropy,liger,backward,speed,ms,BT,B x T,4096,1.1100800037384033,1.1085439920425415,1.116646409034729,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:36:01,0.6.3
+fused_linear_cross_entropy,liger,backward,speed,ms,BT,B x T,8192,1.1324639916419983,1.1306175947189332,1.1396095752716064,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:36:01,0.6.3
+fused_linear_cross_entropy,liger,backward,speed,ms,BT,B x T,16384,1.1767359972000122,1.1753535747528077,1.1848128318786622,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:36:01,0.6.3
+fused_linear_cross_entropy,liger,backward,speed,ms,BT,B x T,32768,1.2614880204200745,1.2595839500427246,1.264524793624878,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:36:01,0.6.3
+fused_linear_cross_entropy,liger-fp32-accum,backward,speed,ms,BT,B x T,4096,1.1116799712181091,1.1089408159255982,1.1186880350112915,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:36:14,0.6.3
+fused_linear_cross_entropy,liger-fp32-accum,backward,speed,ms,BT,B x T,8192,1.1335680484771729,1.1312896013259888,1.137727999687195,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:36:14,0.6.3
+fused_linear_cross_entropy,liger-fp32-accum,backward,speed,ms,BT,B x T,16384,1.1740800142288208,1.172153615951538,1.1790144205093385,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:36:14,0.6.3
+fused_linear_cross_entropy,liger-fp32-accum,backward,speed,ms,BT,B x T,32768,1.2629120349884033,1.2613375425338746,1.268000030517578,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:36:14,0.6.3
+fused_linear_cross_entropy,huggingface,backward,speed,ms,BT,B x T,4096,13.298080444335938,13.274892997741699,13.363347244262696,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:36:28,0.6.3
+fused_linear_cross_entropy,huggingface,backward,speed,ms,BT,B x T,8192,26.327648162841797,26.049958419799804,26.4451904296875,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:36:28,0.6.3
+fused_linear_cross_entropy,huggingface,backward,speed,ms,BT,B x T,16384,52.90752029418945,52.90752029418945,52.90752029418945,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:36:28,0.6.3
+fused_linear_cross_entropy,huggingface,backward,speed,ms,BT,B x T,32768,106.2070083618164,106.2070083618164,106.2070083618164,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:36:28,0.6.3
+fused_linear_cross_entropy,liger,full,speed,ms,BT,B x T,4096,156.70774841308594,156.70774841308594,156.70774841308594,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:36:45,0.6.3
+fused_linear_cross_entropy,liger,full,speed,ms,BT,B x T,8192,161.40582275390625,161.40582275390625,161.40582275390625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:36:45,0.6.3
+fused_linear_cross_entropy,liger,full,speed,ms,BT,B x T,16384,193.07264709472656,193.07264709472656,193.07264709472656,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:36:45,0.6.3
+fused_linear_cross_entropy,liger,full,speed,ms,BT,B x T,32768,269.19580078125,269.19580078125,269.19580078125,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:36:45,0.6.3
+fused_linear_cross_entropy,liger-fp32-accum,full,speed,ms,BT,B x T,4096,157.5746307373047,157.5746307373047,157.5746307373047,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:37:04,0.6.3
+fused_linear_cross_entropy,liger-fp32-accum,full,speed,ms,BT,B x T,8192,162.08995056152344,162.08995056152344,162.08995056152344,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:37:04,0.6.3
+fused_linear_cross_entropy,liger-fp32-accum,full,speed,ms,BT,B x T,16384,193.63734436035156,193.63734436035156,193.63734436035156,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:37:04,0.6.3
+fused_linear_cross_entropy,liger-fp32-accum,full,speed,ms,BT,B x T,32768,263.23358154296875,263.23358154296875,263.23358154296875,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:37:04,0.6.3
+fused_linear_cross_entropy,huggingface,full,speed,ms,BT,B x T,4096,20.66089630126953,20.536934280395506,20.76217613220215,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:37:18,0.6.3
+fused_linear_cross_entropy,huggingface,full,speed,ms,BT,B x T,8192,40.12492752075195,40.07815551757813,40.17169952392578,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:37:18,0.6.3
+fused_linear_cross_entropy,huggingface,full,speed,ms,BT,B x T,16384,85.05133056640625,85.05133056640625,85.05133056640625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:37:18,0.6.3
+fused_linear_cross_entropy,huggingface,full,speed,ms,BT,B x T,32768,168.36904907226562,168.36904907226562,168.36904907226562,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:37:18,0.6.3
+fused_linear_cross_entropy,liger,no-grad-forward,speed,ms,BT,B x T,4096,155.29641723632812,155.29641723632812,155.29641723632812,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:37:36,0.6.3
+fused_linear_cross_entropy,liger,no-grad-forward,speed,ms,BT,B x T,8192,160.10508728027344,160.10508728027344,160.10508728027344,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:37:36,0.6.3
+fused_linear_cross_entropy,liger,no-grad-forward,speed,ms,BT,B x T,16384,191.93557739257812,191.93557739257812,191.93557739257812,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:37:36,0.6.3
+fused_linear_cross_entropy,liger,no-grad-forward,speed,ms,BT,B x T,32768,266.38873291015625,266.38873291015625,266.38873291015625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:37:36,0.6.3
+fused_linear_cross_entropy,liger-fp32-accum,no-grad-forward,speed,ms,BT,B x T,4096,156.4625244140625,156.4625244140625,156.4625244140625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:37:53,0.6.3
+fused_linear_cross_entropy,liger-fp32-accum,no-grad-forward,speed,ms,BT,B x T,8192,160.7310791015625,160.7310791015625,160.7310791015625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:37:53,0.6.3
+fused_linear_cross_entropy,liger-fp32-accum,no-grad-forward,speed,ms,BT,B x T,16384,192.33062744140625,192.33062744140625,192.33062744140625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:37:53,0.6.3
+fused_linear_cross_entropy,liger-fp32-accum,no-grad-forward,speed,ms,BT,B x T,32768,263.2972717285156,263.2972717285156,263.2972717285156,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:37:53,0.6.3
+fused_linear_cross_entropy,huggingface,no-grad-forward,speed,ms,BT,B x T,4096,7.062368154525757,6.887558460235596,7.1631745338439945,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:38:06,0.6.3
+fused_linear_cross_entropy,huggingface,no-grad-forward,speed,ms,BT,B x T,8192,13.76899242401123,13.514348983764648,13.907033348083496,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:38:06,0.6.3
+fused_linear_cross_entropy,huggingface,no-grad-forward,speed,ms,BT,B x T,16384,28.175424575805664,27.54522247314453,28.44234313964844,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:38:06,0.6.3
+fused_linear_cross_entropy,huggingface,no-grad-forward,speed,ms,BT,B x T,32768,54.708351135253906,54.708351135253906,54.708351135253906,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:38:06,0.6.3
+fused_linear_cross_entropy,liger,full,memory,MB,BT,B x T,4096,6203.36328125,6203.36328125,6203.36328125,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:38:25,0.6.3
+fused_linear_cross_entropy,liger,full,memory,MB,BT,B x T,8192,6330.7265625,6330.7265625,6330.7265625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:38:25,0.6.3
+fused_linear_cross_entropy,liger,full,memory,MB,BT,B x T,16384,6585.453125,6585.453125,6585.453125,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:38:25,0.6.3
+fused_linear_cross_entropy,liger,full,memory,MB,BT,B x T,32768,7094.90625,7094.90625,7094.90625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:38:25,0.6.3
+fused_linear_cross_entropy,liger-fp32-accum,full,memory,MB,BT,B x T,4096,7205.36328125,7205.36328125,7205.36328125,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:38:46,0.6.3
+fused_linear_cross_entropy,liger-fp32-accum,full,memory,MB,BT,B x T,8192,7332.7265625,7332.7265625,7332.7265625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:38:46,0.6.3
+fused_linear_cross_entropy,liger-fp32-accum,full,memory,MB,BT,B x T,16384,7587.453125,7587.453125,7587.453125,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:38:46,0.6.3
+fused_linear_cross_entropy,liger-fp32-accum,full,memory,MB,BT,B x T,32768,8096.90625,8096.90625,8096.90625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:38:46,0.6.3
+fused_linear_cross_entropy,huggingface,full,memory,MB,BT,B x T,4096,5138.0322265625,5138.0322265625,5138.0322265625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:38:58,0.6.3
+fused_linear_cross_entropy,huggingface,full,memory,MB,BT,B x T,8192,8208.0634765625,8208.0634765625,8208.0634765625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:38:58,0.6.3
+fused_linear_cross_entropy,huggingface,full,memory,MB,BT,B x T,16384,14348.1259765625,14348.1259765625,14348.1259765625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:38:58,0.6.3
+fused_linear_cross_entropy,huggingface,full,memory,MB,BT,B x T,32768,26628.25,26628.25,26628.25,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:38:58,0.6.3
+fused_linear_jsd,liger,forward,speed,ms,BT,B x T,1024,127.54476928710938,127.54476928710938,127.54476928710938,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:20,0.6.3
+fused_linear_jsd,liger,forward,speed,ms,BT,B x T,2048,148.9003448486328,148.9003448486328,148.9003448486328,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:20,0.6.3
+fused_linear_jsd,liger,forward,speed,ms,BT,B x T,4096,174.50656127929688,174.50656127929688,174.50656127929688,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:20,0.6.3
+fused_linear_jsd,liger,forward,speed,ms,BT,B x T,8192,222.6829071044922,222.6829071044922,222.6829071044922,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:20,0.6.3
+fused_linear_jsd,torch,forward,speed,ms,BT,B x T,1024,9.46460771560669,9.454399871826173,9.467808151245118,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:21,0.6.3
+fused_linear_jsd,torch,forward,speed,ms,BT,B x T,2048,18.614656448364258,18.613459396362305,18.621068954467773,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:21,0.6.3
+fused_linear_jsd,torch,forward,speed,ms,BT,B x T,4096,37.04670333862305,37.03450164794922,37.058905029296874,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:21,0.6.3
+fused_linear_jsd,torch,forward,speed,ms,BT,B x T,8192,73.80313873291016,73.80313873291016,73.80313873291016,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:21,0.6.3
+fused_linear_jsd,liger,backward,speed,ms,BT,B x T,1024,1.0971840023994446,1.09268479347229,1.1036543607711793,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:22,0.6.3
+fused_linear_jsd,liger,backward,speed,ms,BT,B x T,2048,1.098863959312439,1.0951488256454467,1.107820749282837,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:22,0.6.3
+fused_linear_jsd,liger,backward,speed,ms,BT,B x T,4096,1.1099200248718262,1.107200026512146,1.1202239990234375,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:22,0.6.3
+fused_linear_jsd,liger,backward,speed,ms,BT,B x T,8192,1.1303999423980713,1.1282240152359009,1.1384703874588014,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:22,0.6.3
+fused_linear_jsd,torch,backward,speed,ms,BT,B x T,1024,14.386623859405518,14.385248184204102,14.38806438446045,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:24,0.6.3
+fused_linear_jsd,torch,backward,speed,ms,BT,B x T,2048,28.379039764404297,28.378425216674803,28.385298538208012,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:24,0.6.3
+fused_linear_jsd,torch,backward,speed,ms,BT,B x T,4096,56.64409637451172,56.64409637451172,56.64409637451172,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:24,0.6.3
+fused_linear_jsd,torch,backward,speed,ms,BT,B x T,8192,113.76435089111328,113.76435089111328,113.76435089111328,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:24,0.6.3
+fused_linear_jsd,liger,full,speed,ms,BT,B x T,1024,128.97789001464844,128.97789001464844,128.97789001464844,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:30,0.6.3
+fused_linear_jsd,liger,full,speed,ms,BT,B x T,2048,150.2821502685547,150.2821502685547,150.2821502685547,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:30,0.6.3
+fused_linear_jsd,liger,full,speed,ms,BT,B x T,4096,175.1346893310547,175.1346893310547,175.1346893310547,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:30,0.6.3
+fused_linear_jsd,liger,full,speed,ms,BT,B x T,8192,223.9784393310547,223.9784393310547,223.9784393310547,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:30,0.6.3
+fused_linear_jsd,torch,full,speed,ms,BT,B x T,1024,25.8919677734375,25.89079704284668,25.90108757019043,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:33,0.6.3
+fused_linear_jsd,torch,full,speed,ms,BT,B x T,2048,49.11556816101074,49.105017852783206,49.12611846923828,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:33,0.6.3
+fused_linear_jsd,torch,full,speed,ms,BT,B x T,4096,95.86707305908203,95.86707305908203,95.86707305908203,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:33,0.6.3
+fused_linear_jsd,torch,full,speed,ms,BT,B x T,8192,193.23455810546875,193.23455810546875,193.23455810546875,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:33,0.6.3
+fused_linear_jsd,liger,full,memory,MB,BT,B x T,1024,5662.3125,5662.3125,5662.3125,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:40,0.6.3
+fused_linear_jsd,liger,full,memory,MB,BT,B x T,2048,6249.59375,6249.59375,6249.59375,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:40,0.6.3
+fused_linear_jsd,liger,full,memory,MB,BT,B x T,4096,7425.1875,7425.1875,7425.1875,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:40,0.6.3
+fused_linear_jsd,liger,full,memory,MB,BT,B x T,8192,9776.375,9776.375,9776.375,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:40,0.6.3
+fused_linear_jsd,torch,full,memory,MB,BT,B x T,1024,10609.00390625,10609.00390625,10609.00390625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:42,0.6.3
+fused_linear_jsd,torch,full,memory,MB,BT,B x T,2048,17146.0078125,17146.0078125,17146.0078125,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:42,0.6.3
+fused_linear_jsd,torch,full,memory,MB,BT,B x T,4096,30220.015625,30220.015625,30220.015625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:42,0.6.3
+fused_linear_jsd,torch,full,memory,MB,BT,B x T,8192,56368.03125,56368.03125,56368.03125,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:42,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,64,0.35651201009750366,0.3304831922054291,0.37364478707313536,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:55,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,128,0.3453119993209839,0.32655999064445496,0.3662720024585724,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:55,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,256,0.35100799798965454,0.32746880650520327,0.37197439670562743,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:55,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,512,0.36182400584220886,0.34117760062217717,0.3834559917449951,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:55,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,1024,0.35366401076316833,0.3405439853668213,0.3787840008735657,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:55,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,2048,0.7873920202255249,0.7854400277137756,0.7884799838066101,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:55,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,4096,2.6189600229263306,2.597343921661377,2.6197121143341064,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:39:55,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,64,5.951296091079712,5.935296058654785,5.966239929199219,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:02,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,128,11.595344066619873,11.560530853271484,11.624512100219725,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:02,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,256,22.119215965270996,22.091493606567383,22.13814468383789,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:02,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,512,43.58440017700195,43.543206024169926,43.62559432983398,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:02,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,1024,89.80719757080078,89.80719757080078,89.80719757080078,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:02,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,2048,174.71180725097656,174.71180725097656,174.71180725097656,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:02,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,4096,351.46905517578125,351.46905517578125,351.46905517578125,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:02,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,64,1.572543978691101,1.4799168348312377,1.7050687789916994,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:13,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,128,1.5652480125427246,1.5223424196243287,1.6497023820877077,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:13,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,256,1.5531839728355408,1.4921280145645142,1.6464320421218872,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:13,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,512,1.6158080101013184,1.5457919836044312,1.792512059211731,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:13,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,1024,1.6002559661865234,1.5191680192947388,1.6792000532150269,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:13,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,2048,2.6822080612182617,2.678816080093384,2.7090880393981935,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:13,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,4096,8.698176383972168,8.669183731079102,8.703104019165039,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:13,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,64,6.72108793258667,6.669568061828613,6.8514303207397464,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:20,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,128,12.222064018249512,12.152972793579103,12.30444793701172,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:20,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,256,23.528032302856445,23.39098243713379,24.040544509887695,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:20,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,512,44.92753601074219,44.89110488891602,44.96396713256836,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:20,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,1024,87.91423797607422,87.91423797607422,87.91423797607422,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:20,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,2048,180.2018280029297,180.2018280029297,180.2018280029297,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:20,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,4096,358.3681945800781,358.3681945800781,358.3681945800781,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:20,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,64,0.8171839714050293,0.7932544231414795,0.8692031860351563,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:21,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,128,0.8245760202407837,0.8014015793800354,0.8888704180717468,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:21,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,256,0.8463039994239807,0.8112959861755371,0.9121919870376587,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:21,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,512,0.8753760159015656,0.8464960098266602,0.9617279767990112,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:21,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,1024,0.8695840239524841,0.846937608718872,0.909228789806366,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:21,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,2048,1.9208639860153198,1.9193920135498048,1.9221120357513428,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:21,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,4096,6.073696136474609,6.069068813323974,6.0811456680297855,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:21,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,64,0.6738880276679993,0.650662398338318,0.7154175996780396,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:23,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,128,0.7184000015258789,0.677785587310791,0.7689472079277038,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:23,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,256,0.7028800249099731,0.6749696254730225,0.7812735915184021,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:23,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,512,0.743232011795044,0.720736026763916,0.8015999794006348,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:23,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,1024,0.952895998954773,0.9513215780258178,0.9545279741287231,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:23,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,2048,2.5739200115203857,2.571961498260498,2.575551986694336,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:23,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,4096,8.67529582977295,8.673215866088867,8.678879737854004,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:23,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,64,0.3495680093765259,0.3272320032119751,0.37462401390075684,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:29,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,128,0.34886400401592255,0.3290239930152893,0.3726783931255341,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:29,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,256,0.36051198840141296,0.34569600224494934,0.3834240138530731,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:29,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,512,0.42979198694229126,0.42895999550819397,0.4305727958679199,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:29,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,1024,0.985727995634079,0.983135998249054,0.9936512231826782,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:29,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,2048,2.7195520401000977,2.7187007904052733,2.7205952167510983,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:29,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,4096,8.422207832336426,8.418463706970215,8.505632400512695,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:29,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,64,5.621119976043701,5.601791858673096,5.66928653717041,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:35,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,128,11.05020809173584,11.041542243957519,11.098246192932129,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:35,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,256,21.892623901367188,21.784498977661134,21.95023994445801,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:35,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,512,44.24731254577637,44.03301162719727,44.461613464355466,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:35,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,1024,90.70342254638672,90.70342254638672,90.70342254638672,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:35,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,2048,175.81312561035156,175.81312561035156,175.81312561035156,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:35,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,4096,365.462158203125,365.462158203125,365.462158203125,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:35,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,64,1.4821120500564575,1.4450752019882203,1.5544064283370973,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:45,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,128,1.4830080270767212,1.4453824043273926,1.5408191919326784,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:45,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,256,1.5674880146980286,1.5295679569244385,1.634559988975525,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:45,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,512,1.5439679622650146,1.4943104028701784,1.6080895900726317,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:45,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,1024,3.147984027862549,3.1446720600128173,3.152115249633789,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:45,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,2048,8.712544441223145,8.630175590515137,8.717215538024902,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:45,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,4096,27.570335388183594,27.50497932434082,27.57346534729004,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:45,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,64,6.608128070831299,6.58856954574585,6.812671852111817,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:52,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,128,12.381167888641357,12.069433784484865,12.55191650390625,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:52,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,256,23.181663513183594,23.038924407958987,23.200614166259765,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:52,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,512,45.80847930908203,45.71821975708008,45.898738861083984,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:52,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,1024,92.46749114990234,92.46749114990234,92.46749114990234,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:52,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,2048,184.8612823486328,184.8612823486328,184.8612823486328,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:52,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,4096,389.39312744140625,389.39312744140625,389.39312744140625,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:52,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,64,0.8144800066947937,0.7869888067245483,0.8936832189559936,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:53,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,128,0.8103359937667847,0.775161612033844,0.896236789226532,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:53,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,256,0.8589920103549957,0.8288127779960632,0.9816383838653564,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:53,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,512,0.9537919759750366,0.9527040004730224,0.9550975918769836,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:53,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,1024,2.1832640171051025,2.1816832542419435,2.1869632244110107,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:53,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,2048,5.9285759925842285,5.9264960289001465,5.9306559562683105,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:53,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,4096,19.086912155151367,19.080659866333008,19.121445846557616,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:53,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,64,0.7066240012645721,0.6473471879959106,0.791980791091919,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:56,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,128,0.7201600074768066,0.688102388381958,0.8491392254829406,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:56,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,256,0.7383680045604706,0.7059647917747498,0.8505600214004518,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:56,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,512,1.0839679837226868,1.0829503774642943,1.0850687980651856,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:56,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,1024,2.6770238876342773,2.6749183177947997,2.678995132446289,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:56,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,2048,7.877311944961548,7.861664104461671,7.881759929656982,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:56,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,4096,26.467039108276367,26.462239456176757,26.468959426879884,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:40:56,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,64,0.36129599809646606,0.3428736090660095,0.3822272002696991,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:02,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,128,0.3515999913215637,0.33136001229286194,0.37627519965171813,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:02,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,256,0.34700798988342285,0.3268927991390228,0.3686399877071381,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:02,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,512,0.3529919981956482,0.3510400056838989,0.3744575917720795,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:02,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,1024,0.7074399888515472,0.7032960057258606,0.7090880274772644,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:02,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,2048,1.8629120588302612,1.8615039587020874,1.8640639781951904,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:02,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,4096,6.006864070892334,6.004576206207275,6.009280204772949,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:02,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,64,7.251967906951904,7.16382713317871,7.324153709411622,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:11,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,128,14.443151950836182,14.051008224487305,14.554783821105957,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:11,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,256,29.472288131713867,28.83507881164551,29.563199996948242,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:11,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,512,62.664703369140625,62.664703369140625,62.664703369140625,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:11,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,1024,112.54598236083984,112.54598236083984,112.54598236083984,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:11,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,2048,228.651611328125,228.651611328125,228.651611328125,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:11,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,4096,466.4656677246094,466.4656677246094,466.4656677246094,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:11,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,64,1.492959976196289,1.45414400100708,1.5320639610290527,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:20,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,128,1.4528319835662842,1.4199999570846558,1.4968960523605348,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:20,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,256,1.4516160488128662,1.4130560159683228,1.5099200010299683,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:20,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,512,1.5490880012512207,1.5238399505615234,1.5856000185012817,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:20,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,1024,2.367311954498291,2.3517696380615236,2.374323272705078,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:20,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,2048,6.116847991943359,6.109280109405518,6.1202239990234375,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:20,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,4096,19.41164779663086,19.292658615112305,19.418623352050783,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:20,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,64,8.105663776397705,8.044755172729493,8.14517765045166,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:29,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,128,14.991535663604736,14.917119979858398,15.119359970092773,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:29,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,256,29.044960021972656,28.840095901489256,29.07493095397949,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:29,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,512,57.29484939575195,57.29484939575195,57.29484939575195,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:29,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,1024,113.87481689453125,113.87481689453125,113.87481689453125,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:29,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,2048,230.55030822753906,230.55030822753906,230.55030822753906,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:29,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,4096,469.82086181640625,469.82086181640625,469.82086181640625,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:29,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,64,0.8159039914608002,0.788051187992096,0.858355176448822,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:30,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,128,0.8102399706840515,0.7850112080574037,0.8641343951225281,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:30,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,256,0.8267199993133545,0.7892799973487854,0.8922112107276916,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:30,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,512,0.844543993473053,0.8215296030044555,0.880620813369751,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:30,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,1024,1.662656009197235,1.6611839532852173,1.6682560443878174,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:30,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,2048,4.311535835266113,4.300678253173828,4.314982509613038,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:30,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,4096,13.295136451721191,13.293170738220216,13.391475486755372,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:30,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,64,0.6695359945297241,0.6479487895965577,0.6944384098052978,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:33,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,128,0.6661919951438904,0.6405888080596924,0.6899711966514588,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:33,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,256,0.6998240053653717,0.6659135937690734,0.80181120634079,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:33,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,512,0.9161279797554016,0.9150400161743164,0.9174399971961975,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:33,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,1024,1.9767839908599854,1.974508786201477,1.9785856246948241,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:33,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,2048,5.544288158416748,5.535398292541504,5.546457576751709,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:33,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,4096,18.356639862060547,18.344192886352538,18.35847053527832,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:33,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,64,0.3591359853744507,0.34170239567756655,0.382752001285553,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:37,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,128,0.34753599762916565,0.32896000146865845,0.3699775993824005,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:37,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,256,0.33156800270080566,0.31578240394592283,0.36190080642700195,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:37,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,512,0.34463998675346375,0.3282560110092163,0.37513598799705505,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:37,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,1024,0.3656800091266632,0.3445824086666107,0.38808960318565366,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:37,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,2048,0.7871040105819702,0.7858240127563477,0.7882559895515442,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:37,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,4096,2.606608033180237,2.601759910583496,2.624959945678711,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:37,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,64,5.671616077423096,5.661715030670166,5.743628883361817,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:44,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,128,11.31769609451294,11.288313293457032,11.372857856750489,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:44,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,256,22.33291244506836,22.25566711425781,22.469843292236327,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:44,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,512,44.98263931274414,44.73124465942383,45.23403396606445,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:44,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,1024,89.310302734375,89.310302734375,89.310302734375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:44,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,2048,183.25990295410156,183.25990295410156,183.25990295410156,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:44,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,4096,360.4381408691406,360.4381408691406,360.4381408691406,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:44,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,64,1.4774240255355835,1.4543359756469725,1.5166336297988892,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:45,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,128,1.4476799964904785,1.4157504081726073,1.4843711853027344,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:45,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,256,1.4503520131111145,1.4115840196609497,1.5053119659423828,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:45,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,512,1.515936017036438,1.4884096145629884,1.5628031730651855,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:45,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,1024,1.5209280252456665,1.4626879930496217,1.670137643814087,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:45,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,2048,2.714143991470337,2.7108736515045164,2.715647983551025,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:45,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,4096,8.667776107788086,8.660863876342773,8.669088363647461,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:45,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,64,6.614048004150391,6.4940928459167475,6.688313674926758,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:52,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,128,12.09604787826538,11.833683204650878,12.227411079406739,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:52,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,256,23.202655792236328,23.0627010345459,23.33152656555176,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:52,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,512,46.28292655944824,46.18996963500977,46.37588348388672,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:52,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,1024,89.64672088623047,89.64672088623047,89.64672088623047,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:52,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,2048,182.7382049560547,182.7382049560547,182.7382049560547,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:52,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,4096,369.51641845703125,369.51641845703125,369.51641845703125,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:52,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,64,0.8029280006885529,0.7756352066993714,0.8758720278739929,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:53,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,128,0.8272960186004639,0.8015103936195374,0.9022399902343752,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:53,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,256,0.8138879835605621,0.7867648124694824,0.8636288046836853,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:53,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,512,0.8525760173797607,0.8322880268096924,0.895039975643158,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:53,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,1024,0.843999981880188,0.8097407937049865,0.8745855927467346,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:53,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,2048,1.922111988067627,1.9201855897903441,1.9244479894638062,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:53,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,4096,6.0682878494262695,6.06325101852417,6.070470523834229,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:53,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,64,0.6705600023269653,0.6325759887695312,0.7474880218505859,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:55,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,128,0.6801279783248901,0.6491519927978515,0.7582591772079468,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:55,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,256,0.6958560049533844,0.6746240258216858,0.7723519802093506,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:55,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,512,0.7143519818782806,0.6749760031700134,0.774348783493042,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:55,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,1024,0.9500160217285156,0.9486335873603821,0.9515391945838927,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:55,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,2048,2.572160005569458,2.5703551292419435,2.5740352153778074,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:55,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,4096,8.677311897277832,8.67353630065918,8.681376457214355,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:41:55,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,64,0.3412480056285858,0.3163520097732544,0.3648639917373657,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:00,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,128,0.3339200019836426,0.31932159662246706,0.3675519943237305,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:00,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,256,0.3484639972448349,0.3271807968616486,0.3753791987895966,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:00,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,512,0.355679988861084,0.3356288135051727,0.37726719975471495,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:00,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,1024,0.3495360016822815,0.32998399138450624,0.3740096032619476,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:00,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,2048,0.3590399920940399,0.3416063964366913,0.3815680146217346,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:00,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,4096,1.0897599458694458,1.0880575895309448,1.090880036354065,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:00,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,64,5.669600009918213,5.58739824295044,5.706438255310059,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:07,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,128,10.9868803024292,10.872025489807129,11.10501766204834,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:07,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,256,21.96561622619629,21.821267318725585,21.99165458679199,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:07,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,512,43.43340873718262,43.32755889892579,43.539258575439455,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:07,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,1024,88.8776626586914,88.8776626586914,88.8776626586914,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:07,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,2048,174.03897094726562,174.03897094726562,174.03897094726562,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:07,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,4096,347.4023742675781,347.4023742675781,347.4023742675781,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:07,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,64,1.4405759572982788,1.416972827911377,1.5231935739517213,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:15,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,128,1.3938560485839844,1.341062378883362,1.4389503955841065,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:15,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,256,1.4067840576171875,1.3686271905899048,1.4613056421279909,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:15,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,512,1.483839988708496,1.450387215614319,1.6221888065338137,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:15,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,1024,1.4730559587478638,1.4316799640655518,1.514143943786621,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:15,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,2048,1.4936319589614868,1.4539008140563965,1.559788775444031,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:15,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,4096,2.929152011871338,2.9258367061614994,2.931731176376343,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:15,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,64,6.503039836883545,6.3746113777160645,6.641702365875244,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:22,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,128,12.031775951385498,11.978067016601564,12.092333030700683,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:22,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,256,22.79318332672119,22.729504013061522,22.847391510009764,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:22,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,512,44.91468620300293,44.678199768066406,45.15117263793945,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:22,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,1024,88.36630249023438,88.36630249023438,88.36630249023438,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:22,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,2048,174.6591033935547,174.6591033935547,174.6591033935547,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:22,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,4096,352.3074035644531,352.3074035644531,352.3074035644531,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:22,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,64,0.7947199940681458,0.7645824193954467,0.8504000067710877,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:23,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,128,0.8084159791469574,0.7753152012825012,0.9308800101280214,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:23,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,256,0.7806400060653687,0.7528319835662842,0.8331968069076539,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:23,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,512,0.8178560137748718,0.7947775840759277,0.8618559837341309,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:23,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,1024,0.8135679960250854,0.7854784011840821,0.8795008182525634,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:23,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,2048,0.8241440057754517,0.7948992013931274,0.8860543727874757,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:23,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,4096,1.8444160223007202,1.8424639701843262,1.8461120128631592,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:23,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,64,0.6625919938087463,0.6352319836616516,0.7453824162483216,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:25,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,128,0.6716480255126953,0.6345279932022094,0.786188805103302,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:25,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,256,0.658735990524292,0.6377087950706481,0.7296640276908875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:25,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,512,0.6646400094032288,0.6483008146286011,0.7101888060569763,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:25,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,1024,0.7023680210113525,0.6684095859527588,0.8699840188026429,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:25,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,2048,1.0773119926452637,1.0761727809906005,1.0786559581756592,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:25,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,4096,3.5958720445632935,3.5932159423828125,3.597759962081909,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:25,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,64,0.3498559892177582,0.3303295969963074,0.37167359590530397,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:29,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,128,0.3457919955253601,0.32844799757003784,0.36719998717308044,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:29,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,256,0.3423680067062378,0.3228736102581024,0.36939520239830015,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:29,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,512,0.35147200524806976,0.3329855978488922,0.378604793548584,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:29,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,1024,0.36345601081848145,0.3483903884887696,0.38355199098587034,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:29,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,2048,0.91839998960495,0.917248010635376,0.9195136189460755,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:29,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,4096,3.2768640518188477,3.270438480377197,3.2813695907592773,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:29,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,64,5.610176086425781,5.505068969726563,5.63484182357788,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:36,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,128,11.080416202545166,10.99029769897461,11.175302314758301,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:36,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,256,22.028976440429688,21.85801582336426,22.14565162658691,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:36,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,512,43.12019348144531,42.892826843261716,43.34756011962891,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:36,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,1024,86.9784927368164,86.9784927368164,86.9784927368164,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:36,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,2048,173.41941833496094,173.41941833496094,173.41941833496094,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:36,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,4096,351.1776123046875,351.1776123046875,351.1776123046875,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:36,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,64,1.4425119757652283,1.4132416009902955,1.5518080234527587,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:43,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,128,1.4509919881820679,1.4304575681686402,1.5183871746063233,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:43,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,256,1.481872022151947,1.4480895757675172,1.5569408178329467,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:43,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,512,1.5035200119018555,1.464076805114746,1.5851263761520387,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:43,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,1024,1.531008005142212,1.4818623781204223,1.6294463872909546,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:43,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,2048,2.5623040199279785,2.5567872524261475,2.5647552967071534,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:43,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,4096,9.137696266174316,9.132249450683593,9.13966064453125,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:43,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,64,6.6328959465026855,6.586368083953857,6.7082240104675295,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:50,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,128,12.272672176361084,12.060268974304199,12.520256423950196,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:50,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,256,22.9487361907959,22.707404708862306,23.131200027465823,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:50,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,512,44.726688385009766,44.42983779907227,45.02353897094727,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:50,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,1024,90.11196899414062,90.11196899414062,90.11196899414062,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:50,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,2048,174.8981170654297,174.8981170654297,174.8981170654297,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:50,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,4096,361.248779296875,361.248779296875,361.248779296875,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:50,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,64,0.7764800190925598,0.7433280229568482,0.8908735990524295,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:51,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,128,0.7919679880142212,0.7549759864807128,0.8475519895553589,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:51,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,256,0.7963520288467407,0.7629504203796387,0.8355136156082154,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:51,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,512,0.8061280250549316,0.774015986919403,0.8627071738243103,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:51,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,1024,0.8131040036678314,0.7746880173683167,0.8548543810844421,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:51,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,2048,1.6459200382232666,1.6432960033416748,1.6479040384292603,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:51,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,4096,5.872463941574097,5.868000030517578,5.881120204925537,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:51,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,64,0.6541439890861511,0.6310655951499939,0.718131196498871,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:54,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,128,0.6664800047874451,0.6434688210487366,0.7354111909866333,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:54,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,256,0.6854079961776733,0.6667647957801819,0.7373567938804626,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:54,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,512,0.6903200149536133,0.6679872274398804,0.7308735847473145,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:54,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,1024,0.9528639912605286,0.9512320160865784,0.9540351748466491,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:54,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,2048,2.908992052078247,2.9073280811309816,2.9110079288482664,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:54,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,4096,10.516768455505371,10.505554962158204,10.519968223571777,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:54,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,64,0.3335840106010437,0.3153856098651886,0.3640447974205017,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:59,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,128,0.3345920145511627,0.3124415993690491,0.3581375956535339,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:59,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,256,0.3492479920387268,0.32872320413589473,0.3746047973632813,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:59,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,512,0.3592960089445114,0.338591992855072,0.3771840035915375,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:59,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,1024,0.35702401399612427,0.3419904112815857,0.3789824068546295,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:59,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,2048,0.6213919818401337,0.6200960278511047,0.6227520108222961,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:59,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,4096,2.2500160932540894,2.247718381881714,2.2521344661712646,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:42:59,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,64,7.082767963409424,7.050956726074219,7.122995090484619,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:07,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,128,14.0982084274292,13.976268959045411,14.14591999053955,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:07,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,256,27.86947250366211,27.84606704711914,27.88081932067871,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:07,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,512,57.75183868408203,57.75183868408203,57.75183868408203,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:07,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,1024,106.33955383300781,106.33955383300781,106.33955383300781,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:07,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,2048,211.14186096191406,211.14186096191406,211.14186096191406,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:07,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,4096,436.4353942871094,436.4353942871094,436.4353942871094,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:07,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,64,1.3488320112228394,1.2758143663406372,1.4135359764099122,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:14,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,128,1.3629120588302612,1.2626944065093995,1.5627456426620483,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:14,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,256,1.3675679564476013,1.287775993347168,1.5355199575424194,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:14,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,512,1.360703945159912,1.288383960723877,1.465228819847107,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:14,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,1024,1.4227039813995361,1.3380735635757448,1.5318464040756226,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:14,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,2048,1.7420160174369812,1.7392383813858032,1.745740818977356,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:14,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,4096,6.021471977233887,6.016928195953369,6.028192043304443,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:14,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,64,7.550271987915039,7.50133752822876,7.603327846527099,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:22,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,128,14.537792205810547,14.13288288116455,14.846796607971191,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:22,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,256,26.714784622192383,26.698521423339844,27.47508544921875,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:22,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,512,54.93920135498047,54.93920135498047,54.93920135498047,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:22,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,1024,103.9603500366211,103.9603500366211,103.9603500366211,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:22,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,2048,214.47164916992188,214.47164916992188,214.47164916992188,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:22,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,4096,448.09539794921875,448.09539794921875,448.09539794921875,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:22,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,64,0.7889279723167419,0.7666559934616088,0.8160575985908508,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:24,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,128,0.7894719839096069,0.7663871765136718,0.8569600105285644,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:24,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,256,0.7864319980144501,0.7617984056472777,0.8268480181694031,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:24,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,512,0.8225280046463013,0.7985088109970093,0.8790975928306579,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:24,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,1024,0.8262400031089783,0.8020928263664245,0.8697983980178834,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:24,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,2048,1.120848000049591,1.119481611251831,1.122918438911438,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:24,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,4096,3.7758400440216064,3.773484802246094,3.7804671764373783,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:24,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,64,0.6687039732933044,0.6447167873382568,0.7226304054260253,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:27,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,128,0.6635199785232544,0.6403519988059998,0.6969023823738099,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:27,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,256,0.6715520024299622,0.6522240042686462,0.7357760071754456,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:27,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,512,0.7115839719772339,0.6767360210418701,0.8116415858268737,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:27,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,1024,0.718720018863678,0.6916288018226624,0.8190783977508544,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:27,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,2048,1.9918879866600037,1.989408016204834,1.9942400455474854,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:27,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,4096,7.05244779586792,7.0472513198852536,7.055110454559326,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:27,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,64,0.34755200147628784,0.3286080062389374,0.3704319894313812,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:29,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,128,0.3503679931163788,0.32766079902648926,0.373497599363327,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:29,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,256,0.3456000089645386,0.32736000418663025,0.3724159896373749,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:29,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,512,0.3574400097131729,0.33615360260009763,0.3814207911491394,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:29,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,1024,0.35526400804519653,0.3332863986492157,0.3819840013980865,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:29,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,2048,0.3800640106201172,0.3547584056854248,0.4017408072948456,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:29,0.6.3
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,4096,1.090175986289978,1.087225604057312,1.0926463842391967,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:29,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,64,5.610144138336182,5.516480159759522,5.640467357635498,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:36,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,128,11.323712348937988,11.169235229492188,11.77416934967041,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:36,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,256,22.761568069458008,22.472230911254883,22.81733055114746,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:36,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,512,45.245201110839844,45.17993087768555,45.31047134399414,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:36,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,1024,88.08306884765625,88.08306884765625,88.08306884765625,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:36,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,2048,173.42300415039062,173.42300415039062,173.42300415039062,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:36,0.6.3
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,4096,346.5653076171875,346.5653076171875,346.5653076171875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:36,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,64,1.3944159746170044,1.3306560277938844,1.4893759965896607,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:37,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,128,1.4120960235595703,1.3126143932342529,1.5570112228393556,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:37,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,256,1.3772159814834595,1.30545916557312,1.4942272186279297,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:37,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,512,1.410207986831665,1.3492159843444824,1.5412160158157349,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:37,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,1024,1.4001920223236084,1.3121023893356323,1.6131711721420288,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:37,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,2048,1.431935966014862,1.3570879697799683,1.5251200199127197,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:37,0.6.3
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,4096,2.9319519996643066,2.926604890823364,2.936742401123047,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:37,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,64,6.567055940628052,6.420908832550049,6.77056655883789,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:44,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,128,11.984000205993652,11.886803054809569,12.31095027923584,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:44,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,256,23.160256385803223,22.595878982543944,23.564160156249997,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:44,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,512,45.377248764038086,44.999219512939455,45.75527801513672,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:44,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,1024,86.90636444091797,86.90636444091797,86.90636444091797,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:44,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,2048,169.1128387451172,169.1128387451172,169.1128387451172,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:44,0.6.3
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,4096,349.3948059082031,349.3948059082031,349.3948059082031,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:44,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,64,0.755295991897583,0.7193408012390137,0.8111680150032043,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:45,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,128,0.7867200076580048,0.7274624109268188,0.9239552021026611,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:45,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,256,0.7668160200119019,0.7003200054168701,0.8159040212631226,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:45,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,512,0.7968479990959167,0.7352319955825806,0.8459200263023376,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:45,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,1024,0.7547359764575958,0.7118144154548646,0.8075071811676026,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:45,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,2048,0.8076480031013489,0.7617472171783448,0.8689216136932374,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:45,0.6.3
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,4096,1.842848002910614,1.8402816295623778,1.8465536117553711,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:45,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,64,0.6160320043563843,0.5782912135124206,0.724569594860077,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:47,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,128,0.6749119758605957,0.6332224249839783,0.7300992131233215,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:47,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,256,0.6536960005760193,0.6048191785812378,0.7490560054779054,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:47,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,512,0.6632640063762665,0.5913280129432679,0.7272831916809083,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:47,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,1024,0.666047990322113,0.6325952053070069,0.7161215901374817,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:47,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,2048,1.077407956123352,1.0763519763946534,1.0788480043411255,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:47,0.6.3
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,4096,3.5968480110168457,3.595263957977295,3.59932804107666,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:47,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,64,81.275390625,81.275390625,81.275390625,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:47,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,128,86.5234375,86.5234375,86.5234375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:47,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,256,102.0234375,102.0234375,102.0234375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:47,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,512,151.0234375,151.0234375,151.0234375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:47,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,1024,321.0234375,321.0234375,321.0234375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:47,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,2048,949.0234375,949.0234375,949.0234375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:47,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,4096,3357.0234375,3357.0234375,3357.0234375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:47,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,64,81.279296875,81.279296875,81.279296875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:54,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,128,85.5390625,85.5390625,85.5390625,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:54,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,256,102.0859375,102.0859375,102.0859375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:54,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,512,159.2734375,159.2734375,159.2734375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:54,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,1024,370.0234375,370.0234375,370.0234375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:54,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,2048,1177.0234375,1177.0234375,1177.0234375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:54,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,4096,4333.0234375,4333.0234375,4333.0234375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:54,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,64,104.0380859375,104.0380859375,104.0380859375,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:55,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,128,121.78515625,121.78515625,121.78515625,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:55,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,256,167.78515625,167.78515625,167.78515625,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:55,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,512,314.03515625,314.03515625,314.03515625,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:55,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,1024,824.03515625,824.03515625,824.03515625,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:55,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,2048,2708.03515625,2708.03515625,2708.03515625,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:55,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,4096,9932.03515625,9932.03515625,9932.03515625,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:43:55,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,64,104.0419921875,104.0419921875,104.0419921875,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:02,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,128,118.05078125,118.05078125,118.05078125,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:02,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,256,168.34765625,168.34765625,168.34765625,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:02,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,512,338.28515625,338.28515625,338.28515625,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:02,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,1024,969.03515625,969.03515625,969.03515625,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:02,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,2048,3384.03515625,3384.03515625,3384.03515625,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:02,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,4096,12828.03515625,12828.03515625,12828.03515625,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:02,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,64,123.55078125,123.55078125,123.55078125,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:03,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,128,132.046875,132.046875,132.046875,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:03,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,256,163.046875,163.046875,163.046875,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:03,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,512,261.046875,261.046875,261.046875,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:03,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,1024,601.046875,601.046875,601.046875,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:03,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,2048,1857.046875,1857.046875,1857.046875,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:03,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,4096,6673.046875,6673.046875,6673.046875,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:03,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,64,123.5546875,123.5546875,123.5546875,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:12,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,128,131.06640625,131.06640625,131.06640625,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:12,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,256,163.109375,163.109375,163.109375,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:12,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,512,277.296875,277.296875,277.296875,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:12,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,1024,698.046875,698.046875,698.046875,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:12,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,2048,2309.046875,2309.046875,2309.046875,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:12,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,4096,8609.046875,8609.046875,8609.046875,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:12,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,64,81.275390625,81.275390625,81.275390625,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:13,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,128,86.5234375,86.5234375,86.5234375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:13,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,256,102.0234375,102.0234375,102.0234375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:13,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,512,151.0234375,151.0234375,151.0234375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:13,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,1024,321.0234375,321.0234375,321.0234375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:13,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,2048,949.0234375,949.0234375,949.0234375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:13,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,4096,3357.0234375,3357.0234375,3357.0234375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:13,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,64,81.279296875,81.279296875,81.279296875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:20,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,128,85.5390625,85.5390625,85.5390625,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:20,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,256,102.0859375,102.0859375,102.0859375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:20,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,512,159.2734375,159.2734375,159.2734375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:20,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,1024,370.0234375,370.0234375,370.0234375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:20,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,2048,1177.0234375,1177.0234375,1177.0234375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:20,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,4096,4333.0234375,4333.0234375,4333.0234375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:20,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,64,73.1376953125,73.1376953125,73.1376953125,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:20,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,128,75.76171875,75.76171875,75.76171875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:20,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,256,83.51171875,83.51171875,83.51171875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:20,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,512,108.01171875,108.01171875,108.01171875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:20,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,1024,193.01171875,193.01171875,193.01171875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:20,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,2048,507.01171875,507.01171875,507.01171875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:20,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,4096,1711.01171875,1711.01171875,1711.01171875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:20,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,64,73.1416015625,73.1416015625,73.1416015625,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:27,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,128,75.27734375,75.27734375,75.27734375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:27,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,256,83.57421875,83.57421875,83.57421875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:27,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,512,112.26171875,112.26171875,112.26171875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:27,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,1024,218.01171875,218.01171875,218.01171875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:27,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,2048,623.01171875,623.01171875,623.01171875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:27,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,4096,2207.01171875,2207.01171875,2207.01171875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:27,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,64,84.51904296875,84.51904296875,84.51904296875,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:27,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,128,93.142578125,93.142578125,93.142578125,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:27,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,256,116.517578125,116.517578125,116.517578125,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:27,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,512,189.517578125,189.517578125,189.517578125,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:27,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,1024,445.017578125,445.017578125,445.017578125,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:27,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,2048,1386.517578125,1386.517578125,1386.517578125,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:27,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,4096,4998.517578125,4998.517578125,4998.517578125,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:27,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,64,84.52294921875,84.52294921875,84.52294921875,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:34,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,128,91.658203125,91.658203125,91.658203125,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:34,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,256,116.580078125,116.580078125,116.580078125,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:34,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,512,201.767578125,201.767578125,201.767578125,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:34,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,1024,518.017578125,518.017578125,518.017578125,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:34,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,2048,1727.267578125,1727.267578125,1727.267578125,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:34,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,4096,6454.517578125,6454.517578125,6454.517578125,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:34,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,64,94.275390625,94.275390625,94.275390625,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:35,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,128,98.5234375,98.5234375,98.5234375,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:35,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,256,114.0234375,114.0234375,114.0234375,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:35,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,512,163.0234375,163.0234375,163.0234375,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:35,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,1024,333.0234375,333.0234375,333.0234375,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:35,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,2048,961.0234375,961.0234375,961.0234375,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:35,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,4096,3369.0234375,3369.0234375,3369.0234375,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:35,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,64,94.279296875,94.279296875,94.279296875,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:44,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,128,98.041015625,98.041015625,98.041015625,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:44,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,256,114.0859375,114.0859375,114.0859375,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:44,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,512,171.2734375,171.2734375,171.2734375,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:44,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,1024,382.0234375,382.0234375,382.0234375,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:44,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,2048,1189.0234375,1189.0234375,1189.0234375,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:44,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,4096,4345.0234375,4345.0234375,4345.0234375,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:44,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,64,73.1376953125,73.1376953125,73.1376953125,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:44,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,128,75.76171875,75.76171875,75.76171875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:44,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,256,83.51171875,83.51171875,83.51171875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:44,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,512,108.01171875,108.01171875,108.01171875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:44,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,1024,193.01171875,193.01171875,193.01171875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:44,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,2048,507.01171875,507.01171875,507.01171875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:44,0.6.3
+fused_neighborhood_attention,liger,full,memory,MB,seq_len,sequence length,4096,1711.01171875,1711.01171875,1711.01171875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:44,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,64,73.1416015625,73.1416015625,73.1416015625,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:51,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,128,75.27734375,75.27734375,75.27734375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:51,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,256,83.57421875,83.57421875,83.57421875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:51,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,512,112.26171875,112.26171875,112.26171875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:51,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,1024,218.01171875,218.01171875,218.01171875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:51,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,2048,623.01171875,623.01171875,623.01171875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:51,0.6.3
+fused_neighborhood_attention,torch,full,memory,MB,seq_len,sequence length,4096,2207.01171875,2207.01171875,2207.01171875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:44:51,0.6.3
+geglu,liger,full,speed,ms,T,sequence length,1024,9.389823913574219,9.389823913574219,9.389823913574219,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:03,0.6.3
+geglu,liger,full,speed,ms,T,sequence length,2048,21.182111740112305,21.182111740112305,21.182111740112305,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:03,0.6.3
+geglu,liger,full,speed,ms,T,sequence length,4096,39.231807708740234,39.231807708740234,39.231807708740234,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:03,0.6.3
+geglu,liger,full,speed,ms,T,sequence length,8192,77.44902038574219,77.44902038574219,77.44902038574219,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:03,0.6.3
+geglu,huggingface,full,speed,ms,T,sequence length,1024,9.653568267822266,9.653568267822266,9.653568267822266,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:06,0.6.3
+geglu,huggingface,full,speed,ms,T,sequence length,2048,21.129152297973633,21.129152297973633,21.129152297973633,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:06,0.6.3
+geglu,huggingface,full,speed,ms,T,sequence length,4096,40.55430221557617,40.55430221557617,40.55430221557617,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:06,0.6.3
+geglu,huggingface,full,speed,ms,T,sequence length,8192,79.8780517578125,79.8780517578125,79.8780517578125,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:06,0.6.3
+geglu,liger,forward,speed,ms,T,sequence length,1024,2.964319944381714,2.96243839263916,2.9654335975646973,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:09,0.6.3
+geglu,liger,forward,speed,ms,T,sequence length,2048,6.2406721115112305,6.2406721115112305,6.2406721115112305,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:09,0.6.3
+geglu,liger,forward,speed,ms,T,sequence length,4096,13.566847801208496,13.566847801208496,13.566847801208496,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:09,0.6.3
+geglu,liger,forward,speed,ms,T,sequence length,8192,25.59699249267578,25.59699249267578,25.59699249267578,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:09,0.6.3
+geglu,huggingface,forward,speed,ms,T,sequence length,1024,3.161087989807129,3.15886082649231,3.1625087738037108,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:12,0.6.3
+geglu,huggingface,forward,speed,ms,T,sequence length,2048,6.70911979675293,6.70911979675293,6.70911979675293,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:12,0.6.3
+geglu,huggingface,forward,speed,ms,T,sequence length,4096,13.396991729736328,13.396991729736328,13.396991729736328,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:12,0.6.3
+geglu,huggingface,forward,speed,ms,T,sequence length,8192,25.98588752746582,25.98588752746582,25.98588752746582,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:12,0.6.3
+geglu,liger,backward,speed,ms,T,sequence length,1024,6.178624153137207,6.178624153137207,6.178624153137207,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:16,0.6.3
+geglu,liger,backward,speed,ms,T,sequence length,2048,13.641023635864258,13.641023635864258,13.641023635864258,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:16,0.6.3
+geglu,liger,backward,speed,ms,T,sequence length,4096,26.12495994567871,26.12495994567871,26.12495994567871,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:16,0.6.3
+geglu,liger,backward,speed,ms,T,sequence length,8192,51.85478210449219,51.85478210449219,51.85478210449219,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:16,0.6.3
+geglu,huggingface,backward,speed,ms,T,sequence length,1024,6.386240005493164,6.386240005493164,6.386240005493164,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:19,0.6.3
+geglu,huggingface,backward,speed,ms,T,sequence length,2048,13.581024169921875,13.581024169921875,13.581024169921875,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:19,0.6.3
+geglu,huggingface,backward,speed,ms,T,sequence length,4096,26.73027229309082,26.73027229309082,26.73027229309082,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:19,0.6.3
+geglu,huggingface,backward,speed,ms,T,sequence length,8192,52.80742263793945,52.80742263793945,52.80742263793945,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:19,0.6.3
+geglu,liger,full,memory,MB,T,sequence length,1024,1630.0,1630.0,1630.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:22,0.6.3
+geglu,liger,full,memory,MB,T,sequence length,2048,2594.0,2594.0,2594.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:22,0.6.3
+geglu,liger,full,memory,MB,T,sequence length,4096,4522.0,4522.0,4522.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:22,0.6.3
+geglu,liger,full,memory,MB,T,sequence length,8192,8378.0,8378.0,8378.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:22,0.6.3
+geglu,huggingface,full,memory,MB,T,sequence length,1024,2040.0,2040.0,2040.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:25,0.6.3
+geglu,huggingface,full,memory,MB,T,sequence length,2048,3500.0,3500.0,3500.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:25,0.6.3
+geglu,huggingface,full,memory,MB,T,sequence length,4096,6420.0,6420.0,6420.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:25,0.6.3
+geglu,huggingface,full,memory,MB,T,sequence length,8192,12260.0,12260.0,12260.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:25,0.6.3
+geglu,liger,forward,memory,MB,T,sequence length,1024,966.0,966.0,966.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:28,0.6.3
+geglu,liger,forward,memory,MB,T,sequence length,2048,1610.0,1610.0,1610.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:28,0.6.3
+geglu,liger,forward,memory,MB,T,sequence length,4096,2898.0,2898.0,2898.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:28,0.6.3
+geglu,liger,forward,memory,MB,T,sequence length,8192,5474.0,5474.0,5474.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:28,0.6.3
+geglu,huggingface,forward,memory,MB,T,sequence length,1024,1138.0,1138.0,1138.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:30,0.6.3
+geglu,huggingface,forward,memory,MB,T,sequence length,2048,1954.0,1954.0,1954.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:30,0.6.3
+geglu,huggingface,forward,memory,MB,T,sequence length,4096,3586.0,3586.0,3586.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:30,0.6.3
+geglu,huggingface,forward,memory,MB,T,sequence length,8192,6850.0,6850.0,6850.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:30,0.6.3
+geglu,liger,backward,memory,MB,T,sequence length,1024,1630.0,1630.0,1630.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:33,0.6.3
+geglu,liger,backward,memory,MB,T,sequence length,2048,2594.0,2594.0,2594.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:33,0.6.3
+geglu,liger,backward,memory,MB,T,sequence length,4096,4522.0,4522.0,4522.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:33,0.6.3
+geglu,liger,backward,memory,MB,T,sequence length,8192,8378.0,8378.0,8378.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:33,0.6.3
+geglu,huggingface,backward,memory,MB,T,sequence length,1024,2040.0,2040.0,2040.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:36,0.6.3
+geglu,huggingface,backward,memory,MB,T,sequence length,2048,3500.0,3500.0,3500.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:36,0.6.3
+geglu,huggingface,backward,memory,MB,T,sequence length,4096,6420.0,6420.0,6420.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:36,0.6.3
+geglu,huggingface,backward,memory,MB,T,sequence length,8192,12260.0,12260.0,12260.0,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:36,0.6.3
+group_norm,liger,forward,speed,ms,C,num_channels,32,0.021503999829292297,0.021305600181221962,0.02175999991595745,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:44,0.6.3
+group_norm,liger,forward,speed,ms,C,num_channels,64,0.029343999922275543,0.029120000079274178,0.029632000252604485,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:44,0.6.3
+group_norm,liger,forward,speed,ms,C,num_channels,128,0.04435199871659279,0.043935999274253845,0.044895999133586884,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:44,0.6.3
+group_norm,liger,forward,speed,ms,C,num_channels,256,0.07199999690055847,0.07152000069618225,0.07267200201749802,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:44,0.6.3
+group_norm,liger,forward,speed,ms,C,num_channels,512,0.13315199315547943,0.13184000551700592,0.13462400436401367,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:44,0.6.3
+group_norm,liger,forward,speed,ms,C,num_channels,1024,0.25308799743652344,0.25007360577583315,0.2559039890766144,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:44,0.6.3
+group_norm,liger,forward,speed,ms,C,num_channels,2048,0.5092800259590149,0.5072000026702881,0.5110080242156982,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:44,0.6.3
+group_norm,huggingface,forward,speed,ms,C,num_channels,32,0.02879999950528145,0.028575999662280083,0.02908799983561039,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:48,0.6.3
+group_norm,huggingface,forward,speed,ms,C,num_channels,64,0.044096000492572784,0.04390399903059006,0.044415999203920364,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:48,0.6.3
+group_norm,huggingface,forward,speed,ms,C,num_channels,128,0.07763200253248215,0.07715199887752533,0.07798399776220322,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:48,0.6.3
+group_norm,huggingface,forward,speed,ms,C,num_channels,256,0.13884800672531128,0.13849599659442902,0.13952000439167023,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:48,0.6.3
+group_norm,huggingface,forward,speed,ms,C,num_channels,512,0.2622080147266388,0.26175999641418457,0.26265600323677063,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:48,0.6.3
+group_norm,huggingface,forward,speed,ms,C,num_channels,1024,0.5060799717903137,0.5056959986686707,0.5064640045166016,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:48,0.6.3
+group_norm,huggingface,forward,speed,ms,C,num_channels,2048,0.9972479939460754,0.9966400265693665,0.997817611694336,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:48,0.6.3
+group_norm,liger,full,speed,ms,C,num_channels,32,0.25598399341106415,0.23293440639972685,0.2951551973819732,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:52,0.6.3
+group_norm,liger,full,speed,ms,C,num_channels,64,0.2560960054397583,0.22466560304164887,0.32095999121665963,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:52,0.6.3
+group_norm,liger,full,speed,ms,C,num_channels,128,0.2725760042667389,0.23198719918727875,0.34485120177269,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:52,0.6.3
+group_norm,liger,full,speed,ms,C,num_channels,256,0.26710399985313416,0.2250368058681488,0.32394238710403445,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:52,0.6.3
+group_norm,liger,full,speed,ms,C,num_channels,512,0.35888001322746277,0.3568896114826202,0.36071040630340573,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:52,0.6.3
+group_norm,liger,full,speed,ms,C,num_channels,1024,0.6887360215187073,0.6854079961776733,0.6915647983551025,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:52,0.6.3
+group_norm,liger,full,speed,ms,C,num_channels,2048,1.3834559917449951,1.380895972251892,1.3855551958084107,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:52,0.6.3
+group_norm,huggingface,full,speed,ms,C,num_channels,32,0.14735999703407288,0.12737919986248017,0.17907840013504028,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:55,0.6.3
+group_norm,huggingface,full,speed,ms,C,num_channels,64,0.14867199957370758,0.14655999839305878,0.16819840073585512,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:55,0.6.3
+group_norm,huggingface,full,speed,ms,C,num_channels,128,0.22710399329662323,0.22601599991321564,0.22803199291229248,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:55,0.6.3
+group_norm,huggingface,full,speed,ms,C,num_channels,256,0.374752014875412,0.37382400035858154,0.37567999958992004,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:55,0.6.3
+group_norm,huggingface,full,speed,ms,C,num_channels,512,0.6644160151481628,0.663206422328949,0.6653055906295776,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:55,0.6.3
+group_norm,huggingface,full,speed,ms,C,num_channels,1024,1.244096040725708,1.2429120540618896,1.2451519966125488,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:55,0.6.3
+group_norm,huggingface,full,speed,ms,C,num_channels,2048,2.398207902908325,2.3968639373779297,2.399456024169922,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:55,0.6.3
+group_norm,liger,backward,speed,ms,C,num_channels,32,0.11721599847078323,0.1000640019774437,0.15087999403476715,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:58,0.6.3
+group_norm,liger,backward,speed,ms,C,num_channels,64,0.12755200266838074,0.10295680165290833,0.16438399851322175,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:58,0.6.3
+group_norm,liger,backward,speed,ms,C,num_channels,128,0.13521599769592285,0.1137344017624855,0.16502399742603302,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:58,0.6.3
+group_norm,liger,backward,speed,ms,C,num_channels,256,0.12908799946308136,0.1276479959487915,0.14627839922904973,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:58,0.6.3
+group_norm,liger,backward,speed,ms,C,num_channels,512,0.23068800568580627,0.23001599311828613,0.23136000335216522,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:58,0.6.3
+group_norm,liger,backward,speed,ms,C,num_channels,1024,0.4408319890499115,0.4395520091056824,0.44204801321029663,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:58,0.6.3
+group_norm,liger,backward,speed,ms,C,num_channels,2048,0.8783040046691895,0.8772095799446106,0.8791999816894531,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:45:58,0.6.3
+group_norm,huggingface,backward,speed,ms,C,num_channels,32,0.0910400003194809,0.07724159955978394,0.11248639971017838,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,backward,speed,ms,C,num_channels,64,0.1098560020327568,0.10915199667215347,0.11052799969911575,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,backward,speed,ms,C,num_channels,128,0.15273599326610565,0.15212799608707428,0.15331199765205383,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,backward,speed,ms,C,num_channels,256,0.23865599930286407,0.23814399540424347,0.23923200368881226,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,backward,speed,ms,C,num_channels,512,0.40460801124572754,0.403872013092041,0.4053120017051697,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,backward,speed,ms,C,num_channels,1024,0.7407680153846741,0.7400640249252319,0.7414720058441162,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,backward,speed,ms,C,num_channels,2048,1.4041919708251953,1.4034240245819092,1.404863953590393,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,liger,full,memory,MB,C,num_channels,32,40.01171875,40.01171875,40.01171875,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,liger,full,memory,MB,C,num_channels,64,80.01953125,80.01953125,80.01953125,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,liger,full,memory,MB,C,num_channels,128,160.03515625,160.03515625,160.03515625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,liger,full,memory,MB,C,num_channels,256,320.0703125,320.0703125,320.0703125,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,liger,full,memory,MB,C,num_channels,512,640.140625,640.140625,640.140625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,liger,full,memory,MB,C,num_channels,1024,1280.28125,1280.28125,1280.28125,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,liger,full,memory,MB,C,num_channels,2048,2560.5625,2560.5625,2560.5625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,full,memory,MB,C,num_channels,32,40.06640625,40.06640625,40.06640625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,full,memory,MB,C,num_channels,64,80.12890625,80.12890625,80.12890625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,full,memory,MB,C,num_channels,128,160.25390625,160.25390625,160.25390625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,full,memory,MB,C,num_channels,256,320.5078125,320.5078125,320.5078125,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,full,memory,MB,C,num_channels,512,641.015625,641.015625,641.015625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,full,memory,MB,C,num_channels,1024,1282.03125,1282.03125,1282.03125,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,full,memory,MB,C,num_channels,2048,2564.0625,2564.0625,2564.0625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,liger,forward,memory,MB,C,num_channels,32,40.01171875,40.01171875,40.01171875,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,liger,forward,memory,MB,C,num_channels,64,80.01953125,80.01953125,80.01953125,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,liger,forward,memory,MB,C,num_channels,128,160.03515625,160.03515625,160.03515625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,liger,forward,memory,MB,C,num_channels,256,320.0703125,320.0703125,320.0703125,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,liger,forward,memory,MB,C,num_channels,512,640.140625,640.140625,640.140625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,liger,forward,memory,MB,C,num_channels,1024,1280.28125,1280.28125,1280.28125,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,liger,forward,memory,MB,C,num_channels,2048,2560.5625,2560.5625,2560.5625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,forward,memory,MB,C,num_channels,32,40.06640625,40.06640625,40.06640625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,forward,memory,MB,C,num_channels,64,80.12890625,80.12890625,80.12890625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,forward,memory,MB,C,num_channels,128,160.25390625,160.25390625,160.25390625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,forward,memory,MB,C,num_channels,256,320.5078125,320.5078125,320.5078125,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,forward,memory,MB,C,num_channels,512,641.015625,641.015625,641.015625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,forward,memory,MB,C,num_channels,1024,1282.03125,1282.03125,1282.03125,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,forward,memory,MB,C,num_channels,2048,2564.0625,2564.0625,2564.0625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,liger,backward,memory,MB,C,num_channels,32,40.01171875,40.01171875,40.01171875,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,liger,backward,memory,MB,C,num_channels,64,80.01953125,80.01953125,80.01953125,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,liger,backward,memory,MB,C,num_channels,128,160.03515625,160.03515625,160.03515625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,liger,backward,memory,MB,C,num_channels,256,320.0703125,320.0703125,320.0703125,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,liger,backward,memory,MB,C,num_channels,512,640.140625,640.140625,640.140625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,liger,backward,memory,MB,C,num_channels,1024,1280.28125,1280.28125,1280.28125,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,liger,backward,memory,MB,C,num_channels,2048,2560.5625,2560.5625,2560.5625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,backward,memory,MB,C,num_channels,32,40.06640625,40.06640625,40.06640625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,backward,memory,MB,C,num_channels,64,80.12890625,80.12890625,80.12890625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,backward,memory,MB,C,num_channels,128,160.25390625,160.25390625,160.25390625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,backward,memory,MB,C,num_channels,256,320.5078125,320.5078125,320.5078125,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,backward,memory,MB,C,num_channels,512,641.015625,641.015625,641.015625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,backward,memory,MB,C,num_channels,1024,1282.03125,1282.03125,1282.03125,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+group_norm,huggingface,backward,memory,MB,C,num_channels,2048,2564.0625,2564.0625,2564.0625,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:46:01,0.6.3
+jsd,liger,full,memory,MB,V,vocab size,4096,768.00341796875,768.00341796875,768.00341796875,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:29,0.6.3
+jsd,liger,full,memory,MB,V,vocab size,8192,1536.00341796875,1536.00341796875,1536.00341796875,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:29,0.6.3
+jsd,liger,full,memory,MB,V,vocab size,16384,3072.0048828125,3072.0048828125,3072.0048828125,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:29,0.6.3
+jsd,liger,full,memory,MB,V,vocab size,32768,6144.0087890625,6144.0087890625,6144.0087890625,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:29,0.6.3
+jsd,liger,full,memory,MB,V,vocab size,65536,12288.0166015625,12288.0166015625,12288.0166015625,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:29,0.6.3
+jsd,liger,full,memory,MB,V,vocab size,131072,24576.015625,24576.015625,24576.015625,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:29,0.6.3
+jsd,torch,full,memory,MB,V,vocab size,4096,1664.0009765625,1664.0009765625,1664.0009765625,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:30,0.6.3
+jsd,torch,full,memory,MB,V,vocab size,8192,3328.0009765625,3328.0009765625,3328.0009765625,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:30,0.6.3
+jsd,torch,full,memory,MB,V,vocab size,16384,6656.0009765625,6656.0009765625,6656.0009765625,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:30,0.6.3
+jsd,torch,full,memory,MB,V,vocab size,32768,13312.0009765625,13312.0009765625,13312.0009765625,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:30,0.6.3
+jsd,torch,full,memory,MB,V,vocab size,65536,26624.0,26624.0,26624.0,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:30,0.6.3
+jsd,torch,full,memory,MB,V,vocab size,131072,53248.0,53248.0,53248.0,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:30,0.6.3
+jsd,liger,forward,speed,ms,V,vocab size,4096,0.29948800802230835,0.29894399642944336,0.30000001192092896,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:33,0.6.3
+jsd,liger,forward,speed,ms,V,vocab size,8192,0.5928319990634918,0.5902336001396179,0.5952960252761841,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:33,0.6.3
+jsd,liger,forward,speed,ms,V,vocab size,16384,8.315679550170898,8.314080238342285,8.31824016571045,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:33,0.6.3
+jsd,liger,forward,speed,ms,V,vocab size,32768,17.041791915893555,17.03358726501465,17.051826858520506,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:33,0.6.3
+jsd,liger,forward,speed,ms,V,vocab size,65536,34.6823673248291,34.67395706176758,34.69077758789062,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:33,0.6.3
+jsd,liger,forward,speed,ms,V,vocab size,131072,69.57401275634766,69.57401275634766,69.57401275634766,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:33,0.6.3
+jsd,torch,forward,speed,ms,V,vocab size,4096,1.3463680148124695,1.3449471712112429,1.3473983764648438,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:34,0.6.3
+jsd,torch,forward,speed,ms,V,vocab size,8192,2.612351894378662,2.6109951972961425,2.613536071777344,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:34,0.6.3
+jsd,torch,forward,speed,ms,V,vocab size,16384,5.125984191894531,5.124608039855957,5.127347183227539,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:34,0.6.3
+jsd,torch,forward,speed,ms,V,vocab size,32768,10.1626558303833,10.16059513092041,10.166118049621582,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:34,0.6.3
+jsd,torch,forward,speed,ms,V,vocab size,65536,20.23915195465088,20.234982299804688,20.239456558227538,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:34,0.6.3
+jsd,torch,forward,speed,ms,V,vocab size,131072,40.45454406738281,40.4540153503418,40.45507278442383,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:34,0.6.3
+jsd,liger,backward,speed,ms,V,vocab size,4096,0.3245440125465393,0.3216640055179596,0.32972800731658936,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:35,0.6.3
+jsd,liger,backward,speed,ms,V,vocab size,8192,0.5848800241947174,0.582041597366333,0.5879040002822876,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:35,0.6.3
+jsd,liger,backward,speed,ms,V,vocab size,16384,1.200160026550293,1.1993343591690064,1.2013567686080933,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:35,0.6.3
+jsd,liger,backward,speed,ms,V,vocab size,32768,2.5054399967193604,2.504230403900147,2.5082175731658936,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:35,0.6.3
+jsd,liger,backward,speed,ms,V,vocab size,65536,4.971039772033691,4.965030288696289,4.978771305084228,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:35,0.6.3
+jsd,liger,backward,speed,ms,V,vocab size,131072,9.864912033081055,9.85979537963867,9.885952377319336,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:35,0.6.3
+jsd,torch,backward,speed,ms,V,vocab size,4096,1.8369920253753662,1.8361279964447021,1.8381760120391846,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:36,0.6.3
+jsd,torch,backward,speed,ms,V,vocab size,8192,3.594464063644409,3.593075180053711,3.5960575580596927,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:36,0.6.3
+jsd,torch,backward,speed,ms,V,vocab size,16384,7.194943904876709,7.192345428466798,7.195679950714111,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:36,0.6.3
+jsd,torch,backward,speed,ms,V,vocab size,32768,14.477935791015625,14.473312377929688,14.481696128845215,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:36,0.6.3
+jsd,torch,backward,speed,ms,V,vocab size,65536,28.875904083251953,28.87488670349121,28.876460266113284,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:36,0.6.3
+jsd,torch,backward,speed,ms,V,vocab size,131072,57.69718551635742,57.69718551635742,57.69718551635742,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:36,0.6.3
+jsd,liger,full,speed,ms,V,vocab size,4096,0.6229760050773621,0.6184832096099854,0.6319424033164978,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:38,0.6.3
+jsd,liger,full,speed,ms,V,vocab size,8192,1.180832028388977,1.1774272203445435,1.1864831924438477,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:38,0.6.3
+jsd,liger,full,speed,ms,V,vocab size,16384,9.52291202545166,9.515539550781249,9.530034828186036,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:38,0.6.3
+jsd,liger,full,speed,ms,V,vocab size,32768,19.5644474029541,19.552019119262695,19.57342643737793,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:38,0.6.3
+jsd,liger,full,speed,ms,V,vocab size,65536,39.67936134338379,39.67761383056641,39.68110885620117,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:38,0.6.3
+jsd,liger,full,speed,ms,V,vocab size,131072,79.43427276611328,79.43427276611328,79.43427276611328,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:38,0.6.3
+jsd,torch,full,speed,ms,V,vocab size,4096,3.177824020385742,3.1763647556304933,3.179551982879639,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:40,0.6.3
+jsd,torch,full,speed,ms,V,vocab size,8192,6.195040225982666,6.1923390388488775,6.198285007476807,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:40,0.6.3
+jsd,torch,full,speed,ms,V,vocab size,16384,12.313135623931885,12.311065483093262,12.314655685424803,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:40,0.6.3
+jsd,torch,full,speed,ms,V,vocab size,32768,24.630352020263672,24.62902374267578,24.631257629394533,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:40,0.6.3
+jsd,torch,full,speed,ms,V,vocab size,65536,49.10152053833008,49.10047454833985,49.10256652832031,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:40,0.6.3
+jsd,torch,full,speed,ms,V,vocab size,131072,98.14476776123047,98.14476776123047,98.14476776123047,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:40,0.6.3
+kl_div,liger,full,memory,MB,V,vocab size,4096,384.0009765625,384.0009765625,384.0009765625,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:46,0.6.3
+kl_div,liger,full,memory,MB,V,vocab size,8192,768.0009765625,768.0009765625,768.0009765625,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:46,0.6.3
+kl_div,liger,full,memory,MB,V,vocab size,16384,1536.0009765625,1536.0009765625,1536.0009765625,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:46,0.6.3
+kl_div,liger,full,memory,MB,V,vocab size,32768,3072.0009765625,3072.0009765625,3072.0009765625,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:46,0.6.3
+kl_div,liger,full,memory,MB,V,vocab size,65536,6144.0009765625,6144.0009765625,6144.0009765625,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:46,0.6.3
+kl_div,liger,full,memory,MB,V,vocab size,131072,12288.0009765625,12288.0009765625,12288.0009765625,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:46,0.6.3
+kl_div,torch,full,memory,MB,V,vocab size,4096,448.0,448.0,448.0,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:47,0.6.3
+kl_div,torch,full,memory,MB,V,vocab size,8192,896.0,896.0,896.0,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:47,0.6.3
+kl_div,torch,full,memory,MB,V,vocab size,16384,1792.0,1792.0,1792.0,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:47,0.6.3
+kl_div,torch,full,memory,MB,V,vocab size,32768,3584.0,3584.0,3584.0,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:47,0.6.3
+kl_div,torch,full,memory,MB,V,vocab size,65536,7168.0,7168.0,7168.0,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:47,0.6.3
+kl_div,torch,full,memory,MB,V,vocab size,131072,14336.0,14336.0,14336.0,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:47,0.6.3
+kl_div,liger,forward,speed,ms,V,vocab size,4096,0.06771200150251389,0.06675200164318085,0.06908799707889557,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:48,0.6.3
+kl_div,liger,forward,speed,ms,V,vocab size,8192,0.10931199789047241,0.10841599851846695,0.11075200140476227,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:48,0.6.3
+kl_div,liger,forward,speed,ms,V,vocab size,16384,0.1927040070295334,0.1918720006942749,0.19395199418067932,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:48,0.6.3
+kl_div,liger,forward,speed,ms,V,vocab size,32768,0.3591360002756119,0.3579840064048767,0.36057600378990173,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:48,0.6.3
+kl_div,liger,forward,speed,ms,V,vocab size,65536,0.6919680237770081,0.6905663847923279,0.693561589717865,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:48,0.6.3
+kl_div,liger,forward,speed,ms,V,vocab size,131072,1.360416054725647,1.3561407804489136,1.3663424253463745,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:48,0.6.3
+kl_div,torch,forward,speed,ms,V,vocab size,4096,0.22937600314617157,0.22889600694179535,0.2301120012998581,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:48,0.6.3
+kl_div,torch,forward,speed,ms,V,vocab size,8192,0.42579200863838196,0.4251584053039551,0.4264639914035797,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:48,0.6.3
+kl_div,torch,forward,speed,ms,V,vocab size,16384,0.8151999711990356,0.8145920038223267,0.8158720135688782,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:48,0.6.3
+kl_div,torch,forward,speed,ms,V,vocab size,32768,1.5930559635162354,1.5918400287628174,1.593894362449646,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:48,0.6.3
+kl_div,torch,forward,speed,ms,V,vocab size,65536,3.1495200395584106,3.1482879161834716,3.150918436050415,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:48,0.6.3
+kl_div,torch,forward,speed,ms,V,vocab size,131072,6.259424209594727,6.258112049102783,6.262291145324707,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:48,0.6.3
+kl_div,liger,backward,speed,ms,V,vocab size,4096,0.2924480140209198,0.28872960805892944,0.2983872056007385,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:49,0.6.3
+kl_div,liger,backward,speed,ms,V,vocab size,8192,0.5120959877967834,0.5083519816398621,0.5167679786682129,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:49,0.6.3
+kl_div,liger,backward,speed,ms,V,vocab size,16384,0.9972479939460754,0.9949439764022827,0.9999359846115112,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:49,0.6.3
+kl_div,liger,backward,speed,ms,V,vocab size,32768,2.0084800720214844,2.006278324127197,2.011232089996338,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:49,0.6.3
+kl_div,liger,backward,speed,ms,V,vocab size,65536,3.964095950126648,3.960780715942383,3.966585636138916,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:49,0.6.3
+kl_div,liger,backward,speed,ms,V,vocab size,131072,7.866847991943359,7.863091278076172,7.871206188201905,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:49,0.6.3
+kl_div,torch,backward,speed,ms,V,vocab size,4096,0.2571519911289215,0.2563199996948242,0.2577280104160309,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:50,0.6.3
+kl_div,torch,backward,speed,ms,V,vocab size,8192,0.4910399913787842,0.4901120066642761,0.49167999625205994,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:50,0.6.3
+kl_div,torch,backward,speed,ms,V,vocab size,16384,1.006608009338379,1.0058111667633056,1.0071744441986084,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:50,0.6.3
+kl_div,torch,backward,speed,ms,V,vocab size,32768,2.0767840147018433,2.075648069381714,2.0776000022888184,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:50,0.6.3
+kl_div,torch,backward,speed,ms,V,vocab size,65536,4.14796781539917,4.146048069000244,4.148902416229248,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:50,0.6.3
+kl_div,torch,backward,speed,ms,V,vocab size,131072,8.271615982055664,8.264960289001465,8.274399757385254,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:50,0.6.3
+kl_div,liger,full,speed,ms,V,vocab size,4096,0.3596480041742325,0.3495360016822815,0.39526399970054626,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:51,0.6.3
+kl_div,liger,full,speed,ms,V,vocab size,8192,0.6159040033817291,0.6109439730644226,0.62433922290802,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:51,0.6.3
+kl_div,liger,full,speed,ms,V,vocab size,16384,1.1860159635543823,1.1815296173095704,1.1899776220321654,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:51,0.6.3
+kl_div,liger,full,speed,ms,V,vocab size,32768,2.3639200925827026,2.3620479106903076,2.365881586074829,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:51,0.6.3
+kl_div,liger,full,speed,ms,V,vocab size,65536,4.653151988983154,4.651360034942627,4.658815860748291,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:51,0.6.3
+kl_div,liger,full,speed,ms,V,vocab size,131072,9.218976020812988,9.216806602478027,9.228909111022949,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:51,0.6.3
+kl_div,torch,full,speed,ms,V,vocab size,4096,0.4798080027103424,0.4790784060955048,0.48043520450592037,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:52,0.6.3
+kl_div,torch,full,speed,ms,V,vocab size,8192,0.9139039814472198,0.9133440256118774,0.9146751880645752,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:52,0.6.3
+kl_div,torch,full,speed,ms,V,vocab size,16384,1.8178560137748718,1.8169855833053588,1.8185024499893188,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:52,0.6.3
+kl_div,torch,full,speed,ms,V,vocab size,32768,3.667248010635376,3.666271924972534,3.668287992477417,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:52,0.6.3
+kl_div,torch,full,speed,ms,V,vocab size,65536,7.295423984527588,7.295020675659179,7.2964288711547844,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:52,0.6.3
+kl_div,torch,full,speed,ms,V,vocab size,131072,14.524208068847656,14.522879600524902,14.527008056640625,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-11-21 00:47:52,0.6.3
+kto_loss,liger,forward,speed,ms,B,Batch Size (B),2,3.9768319129943848,3.972908687591553,3.978483200073242,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:48:28,0.6.3
+kto_loss,liger,forward,speed,ms,B,Batch Size (B),4,7.797215938568115,7.791520118713379,7.807040214538574,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:48:28,0.6.3
+kto_loss,liger,forward,speed,ms,B,Batch Size (B),8,15.560224056243896,15.52137565612793,15.626432418823242,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:48:28,0.6.3
+kto_loss,liger,forward,speed,ms,B,Batch Size (B),16,31.10099220275879,31.09974365234375,31.109613037109376,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:48:28,0.6.3
+kto_loss,liger,forward,speed,ms,B,Batch Size (B),32,62.199520111083984,62.199520111083984,62.199520111083984,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:48:28,0.6.3
+kto_loss,huggingface,forward,speed,ms,B,Batch Size (B),2,3.725152015686035,3.7156479358673096,3.739487886428833,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:48:43,0.6.3
+kto_loss,huggingface,forward,speed,ms,B,Batch Size (B),4,6.750256061553955,6.718662261962891,6.7670464515686035,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:48:43,0.6.3
+kto_loss,huggingface,forward,speed,ms,B,Batch Size (B),8,12.573887825012207,12.553209877014162,12.584269142150879,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:48:43,0.6.3
+kto_loss,huggingface,forward,speed,ms,B,Batch Size (B),16,24.29806423187256,24.277817535400388,24.30649604797363,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:48:43,0.6.3
+kto_loss,huggingface,forward,speed,ms,B,Batch Size (B),32,47.827646255493164,47.81437911987305,47.840913391113276,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:48:43,0.6.3
+kto_loss,liger,backward,speed,ms,B,Batch Size (B),2,0.3099200129508972,0.3070720136165619,0.3152064085006714,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:48:57,0.6.3
+kto_loss,liger,backward,speed,ms,B,Batch Size (B),4,0.31033599376678467,0.30710399746894834,0.31768319606781004,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:48:57,0.6.3
+kto_loss,liger,backward,speed,ms,B,Batch Size (B),8,0.3092159926891327,0.30671998858451843,0.3165759921073914,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:48:57,0.6.3
+kto_loss,liger,backward,speed,ms,B,Batch Size (B),16,0.3094079941511154,0.30658560395240786,0.31651840209960935,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:48:57,0.6.3
+kto_loss,liger,backward,speed,ms,B,Batch Size (B),32,0.3091200143098831,0.3061183929443359,0.31614079475402834,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:48:57,0.6.3
+kto_loss,huggingface,backward,speed,ms,B,Batch Size (B),2,2.087104082107544,2.080543947219849,2.095859241485596,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:49:12,0.6.3
+kto_loss,huggingface,backward,speed,ms,B,Batch Size (B),4,3.5737600326538086,3.5664000511169434,3.584575891494751,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:49:12,0.6.3
+kto_loss,huggingface,backward,speed,ms,B,Batch Size (B),8,6.599135875701904,6.586547183990479,6.618003177642823,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:49:12,0.6.3
+kto_loss,huggingface,backward,speed,ms,B,Batch Size (B),16,12.585887908935547,12.565612411499023,12.600518226623535,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:49:12,0.6.3
+kto_loss,huggingface,backward,speed,ms,B,Batch Size (B),32,24.509936332702637,24.50292549133301,24.518189239501954,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:49:12,0.6.3
+kto_loss,liger,full,speed,ms,B,Batch Size (B),2,5.837935924530029,5.672255992889404,5.871200084686279,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:49:27,0.6.3
+kto_loss,liger,full,speed,ms,B,Batch Size (B),4,9.968640327453613,9.827251052856445,10.155327987670898,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:49:27,0.6.3
+kto_loss,liger,full,speed,ms,B,Batch Size (B),8,17.981216430664062,17.880818557739257,18.031007766723633,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:49:27,0.6.3
+kto_loss,liger,full,speed,ms,B,Batch Size (B),16,33.34419250488281,33.25097579956055,33.59269866943359,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:49:27,0.6.3
+kto_loss,liger,full,speed,ms,B,Batch Size (B),32,65.31417846679688,65.31417846679688,65.31417846679688,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:49:27,0.6.3
+kto_loss,huggingface,full,speed,ms,B,Batch Size (B),2,6.208608150482178,6.1360321044921875,6.329100799560548,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:49:42,0.6.3
+kto_loss,huggingface,full,speed,ms,B,Batch Size (B),4,10.481696128845215,10.436710548400878,10.756018829345702,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:49:42,0.6.3
+kto_loss,huggingface,full,speed,ms,B,Batch Size (B),8,19.207679748535156,19.20088996887207,19.259135818481447,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:49:42,0.6.3
+kto_loss,huggingface,full,speed,ms,B,Batch Size (B),16,36.86809539794922,36.81577529907227,36.92041549682617,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:49:42,0.6.3
+kto_loss,huggingface,full,speed,ms,B,Batch Size (B),32,72.4271011352539,72.4271011352539,72.4271011352539,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:49:42,0.6.3
+kto_loss,liger,full,memory,MB,B,Batch Size (B),2,2075.99169921875,2075.99169921875,2075.99169921875,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:04,0.6.3
+kto_loss,liger,full,memory,MB,B,Batch Size (B),4,2081.99951171875,2081.99951171875,2081.99951171875,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:04,0.6.3
+kto_loss,liger,full,memory,MB,B,Batch Size (B),8,2094.01513671875,2094.01513671875,2094.01513671875,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:04,0.6.3
+kto_loss,liger,full,memory,MB,B,Batch Size (B),16,2118.04638671875,2118.04638671875,2118.04638671875,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:04,0.6.3
+kto_loss,liger,full,memory,MB,B,Batch Size (B),32,2165.35888671875,2165.35888671875,2165.35888671875,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:04,0.6.3
+kto_loss,huggingface,full,memory,MB,B,Batch Size (B),2,3078.2451171875,3078.2451171875,3078.2451171875,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:21,0.6.3
+kto_loss,huggingface,full,memory,MB,B,Batch Size (B),4,4834.26123046875,4834.26123046875,4834.26123046875,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:21,0.6.3
+kto_loss,huggingface,full,memory,MB,B,Batch Size (B),8,8348.294921875,8348.294921875,8348.294921875,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:21,0.6.3
+kto_loss,huggingface,full,memory,MB,B,Batch Size (B),16,15377.861328125,15377.861328125,15377.861328125,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:21,0.6.3
+kto_loss,huggingface,full,memory,MB,B,Batch Size (B),32,29437.994140625,29437.994140625,29437.994140625,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:21,0.6.3
+layer_norm,liger,forward,speed,ms,N,hidden size,1024,0.018592000007629395,0.01817600056529045,0.02072319947183132,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:31,0.6.3
+layer_norm,liger,forward,speed,ms,N,hidden size,2048,0.029311999678611755,0.028991999104619026,0.030047999694943428,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:31,0.6.3
+layer_norm,liger,forward,speed,ms,N,hidden size,4096,0.05132799968123436,0.05071999877691269,0.05209600180387497,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:31,0.6.3
+layer_norm,liger,forward,speed,ms,N,hidden size,8192,0.09596800059080124,0.095551997423172,0.0963200032711029,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:31,0.6.3
+layer_norm,liger,forward,speed,ms,N,hidden size,16384,0.18592000007629395,0.1852159947156906,0.18639999628067017,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:31,0.6.3
+layer_norm,huggingface,forward,speed,ms,N,hidden size,1024,0.023072000592947006,0.022816000506281853,0.023360000923275948,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:34,0.6.3
+layer_norm,huggingface,forward,speed,ms,N,hidden size,2048,0.036479998379945755,0.036159999668598175,0.03702399879693985,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:34,0.6.3
+layer_norm,huggingface,forward,speed,ms,N,hidden size,4096,0.06870400160551071,0.06800000369548798,0.06931199878454208,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:34,0.6.3
+layer_norm,huggingface,forward,speed,ms,N,hidden size,8192,0.14803199470043182,0.14732800424098969,0.14870400726795197,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:34,0.6.3
+layer_norm,huggingface,forward,speed,ms,N,hidden size,16384,0.291456013917923,0.29078400135040283,0.2921600043773651,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:34,0.6.3
+layer_norm,liger,backward,speed,ms,N,hidden size,1024,0.22217600047588348,0.17115519642829896,0.3339135944843292,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:38,0.6.3
+layer_norm,liger,backward,speed,ms,N,hidden size,2048,0.18518400192260742,0.15806719958782195,0.2658495962619782,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:38,0.6.3
+layer_norm,liger,backward,speed,ms,N,hidden size,4096,0.19219200313091278,0.18969599902629852,0.23862400650978088,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:38,0.6.3
+layer_norm,liger,backward,speed,ms,N,hidden size,8192,0.31625598669052124,0.3139840066432953,0.31935998797416687,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:38,0.6.3
+layer_norm,liger,backward,speed,ms,N,hidden size,16384,0.734112024307251,0.7317824006080628,0.7365120053291321,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:38,0.6.3
+layer_norm,huggingface,backward,speed,ms,N,hidden size,1024,0.06595200300216675,0.060095999389886856,0.08410239964723587,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:40,0.6.3
+layer_norm,huggingface,backward,speed,ms,N,hidden size,2048,0.09942399710416794,0.09804800152778625,0.10096000134944916,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:40,0.6.3
+layer_norm,huggingface,backward,speed,ms,N,hidden size,4096,0.17856000363826752,0.17759999632835388,0.17945599555969238,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:40,0.6.3
+layer_norm,huggingface,backward,speed,ms,N,hidden size,8192,0.33260801434516907,0.33161601424217224,0.33361280560493467,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:40,0.6.3
+layer_norm,huggingface,backward,speed,ms,N,hidden size,16384,0.6394559741020203,0.6382079720497131,0.6406784176826477,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:40,0.6.3
+layer_norm,liger,full,speed,ms,N,hidden size,1024,0.28617599606513977,0.2573375999927521,0.40328959822654725,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:42,0.6.3
+layer_norm,liger,full,speed,ms,N,hidden size,2048,0.28883200883865356,0.2464832067489624,0.38183040022850034,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:42,0.6.3
+layer_norm,liger,full,speed,ms,N,hidden size,4096,0.27775999903678894,0.24092799425125122,0.32550400495529175,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:42,0.6.3
+layer_norm,liger,full,speed,ms,N,hidden size,8192,0.4105280041694641,0.4086399972438812,0.4135807931423187,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:42,0.6.3
+layer_norm,liger,full,speed,ms,N,hidden size,16384,0.9192320108413696,0.9165824174880981,0.9215296268463136,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:42,0.6.3
+layer_norm,huggingface,full,speed,ms,N,hidden size,1024,0.10451199859380722,0.09041280299425125,0.12635519802570344,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:44,0.6.3
+layer_norm,huggingface,full,speed,ms,N,hidden size,2048,0.13116799294948578,0.12963199615478516,0.1329279989004135,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:44,0.6.3
+layer_norm,huggingface,full,speed,ms,N,hidden size,4096,0.24294400215148926,0.24188800156116486,0.2438720017671585,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:44,0.6.3
+layer_norm,huggingface,full,speed,ms,N,hidden size,8192,0.4766719937324524,0.47542399168014526,0.4777855932712555,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:44,0.6.3
+layer_norm,huggingface,full,speed,ms,N,hidden size,16384,0.9275839924812317,0.9263039827346802,0.9290751934051513,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:44,0.6.3
+layer_norm,liger,full,memory,MB,N,hidden size,1024,80.0625,80.0625,80.0625,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:44,0.6.3
+layer_norm,liger,full,memory,MB,N,hidden size,2048,160.09375,160.09375,160.09375,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:44,0.6.3
+layer_norm,liger,full,memory,MB,N,hidden size,4096,320.15625,320.15625,320.15625,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:44,0.6.3
+layer_norm,liger,full,memory,MB,N,hidden size,8192,640.28125,640.28125,640.28125,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:44,0.6.3
+layer_norm,liger,full,memory,MB,N,hidden size,16384,1280.53125,1280.53125,1280.53125,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:44,0.6.3
+layer_norm,huggingface,full,memory,MB,N,hidden size,1024,80.0625,80.0625,80.0625,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:44,0.6.3
+layer_norm,huggingface,full,memory,MB,N,hidden size,2048,160.09375,160.09375,160.09375,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:44,0.6.3
+layer_norm,huggingface,full,memory,MB,N,hidden size,4096,320.15625,320.15625,320.15625,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:44,0.6.3
+layer_norm,huggingface,full,memory,MB,N,hidden size,8192,640.28125,640.28125,640.28125,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:44,0.6.3
+layer_norm,huggingface,full,memory,MB,N,hidden size,16384,1280.53125,1280.53125,1280.53125,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:44,0.6.3
+llama4_rope,liger,forward,speed,ms,H,hidden size,512,0.06230400130152702,0.062111999839544296,0.06252799928188324,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:54,0.6.3
+llama4_rope,liger,forward,speed,ms,H,hidden size,2048,0.06255999952554703,0.062272001057863235,0.06284800171852112,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:54,0.6.3
+llama4_rope,liger,forward,speed,ms,H,hidden size,8192,0.06931199878454208,0.06864000111818314,0.07126399874687195,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:54,0.6.3
+llama4_rope,huggingface,forward,speed,ms,H,hidden size,512,0.031936001032590866,0.030880000442266464,0.03433600068092346,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:55,0.6.3
+llama4_rope,huggingface,forward,speed,ms,H,hidden size,2048,0.0636960007250309,0.06224000081419945,0.0663359984755516,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:55,0.6.3
+llama4_rope,huggingface,forward,speed,ms,H,hidden size,8192,0.20908799767494202,0.20732800662517548,0.21142399311065674,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:55,0.6.3
+llama4_rope,liger,backward,speed,ms,H,hidden size,512,0.1648319959640503,0.14825600385665894,0.20455680489540104,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:56,0.6.3
+llama4_rope,liger,backward,speed,ms,H,hidden size,2048,0.16432000696659088,0.14789119958877564,0.2050175964832306,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:56,0.6.3
+llama4_rope,liger,backward,speed,ms,H,hidden size,8192,0.17110399901866913,0.14941440522670746,0.21392640173435218,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:56,0.6.3
+llama4_rope,huggingface,backward,speed,ms,H,hidden size,512,0.1841920018196106,0.16631680428981782,0.2151423990726471,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:57,0.6.3
+llama4_rope,huggingface,backward,speed,ms,H,hidden size,2048,0.18281599879264832,0.16375039517879486,0.2145279973745346,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:57,0.6.3
+llama4_rope,huggingface,backward,speed,ms,H,hidden size,8192,0.21452799439430237,0.21270400285720825,0.21768959760665893,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:57,0.6.3
+llama4_rope,liger,full,speed,ms,H,hidden size,512,0.3189920037984848,0.29783040285110474,0.35038080215454104,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:58,0.6.3
+llama4_rope,liger,full,speed,ms,H,hidden size,2048,0.3176960051059723,0.29219840168952943,0.35393919944763186,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:58,0.6.3
+llama4_rope,liger,full,speed,ms,H,hidden size,8192,0.31721600890159607,0.2923328042030334,0.3479296028614044,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:50:58,0.6.3
+llama4_rope,huggingface,full,speed,ms,H,hidden size,512,0.29500800371170044,0.27283841371536255,0.32080639600753785,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:00,0.6.3
+llama4_rope,huggingface,full,speed,ms,H,hidden size,2048,0.29742400348186493,0.2758911967277527,0.32455680370330814,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:00,0.6.3
+llama4_rope,huggingface,full,speed,ms,H,hidden size,8192,0.41734400391578674,0.4158911943435669,0.4216639995574951,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:00,0.6.3
+llama4_rope,liger,full,memory,MB,H,hidden size,512,37.23486328125,37.23486328125,37.23486328125,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:00,0.6.3
+llama4_rope,liger,full,memory,MB,H,hidden size,2048,52.89111328125,52.89111328125,52.89111328125,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:00,0.6.3
+llama4_rope,liger,full,memory,MB,H,hidden size,8192,115.51611328125,115.51611328125,115.51611328125,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:00,0.6.3
+llama4_rope,huggingface,full,memory,MB,H,hidden size,512,49.64111328125,49.64111328125,49.64111328125,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:00,0.6.3
+llama4_rope,huggingface,full,memory,MB,H,hidden size,2048,102.51611328125,102.51611328125,102.51611328125,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:00,0.6.3
+llama4_rope,huggingface,full,memory,MB,H,hidden size,8192,314.01611328125,314.01611328125,314.01611328125,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:00,0.6.3
+llama4_rope,liger,forward,speed,ms,T,sequence length,1024,0.04499199986457825,0.04419200122356415,0.0498879998922348,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:02,0.6.3
+llama4_rope,liger,forward,speed,ms,T,sequence length,2048,0.06883200258016586,0.06844799965620041,0.0692799985408783,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:02,0.6.3
+llama4_rope,liger,forward,speed,ms,T,sequence length,4096,0.11680000275373459,0.11644800007343292,0.11711999773979187,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:02,0.6.3
+llama4_rope,liger,forward,speed,ms,T,sequence length,8192,0.2165759950876236,0.2160319983959198,0.21711360514163972,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:02,0.6.3
+llama4_rope,liger,forward,speed,ms,T,sequence length,16384,0.41679999232292175,0.41625601053237915,0.4174720048904419,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:02,0.6.3
+llama4_rope,huggingface,forward,speed,ms,T,sequence length,1024,0.1173119992017746,0.11670400202274323,0.11795199662446976,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:04,0.6.3
+llama4_rope,huggingface,forward,speed,ms,T,sequence length,2048,0.20723199844360352,0.20662400126457214,0.20790399610996246,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:04,0.6.3
+llama4_rope,huggingface,forward,speed,ms,T,sequence length,4096,0.38659200072288513,0.3856959939002991,0.38730879426002507,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:04,0.6.3
+llama4_rope,huggingface,forward,speed,ms,T,sequence length,8192,0.7413599789142609,0.7407616138458253,0.7421311855316162,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:04,0.6.3
+llama4_rope,huggingface,forward,speed,ms,T,sequence length,16384,1.4540159702301025,1.4530880451202393,1.455180788040161,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:04,0.6.3
+llama4_rope,liger,backward,speed,ms,T,sequence length,1024,0.18246400356292725,0.16630400717258453,0.22038400173187256,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:06,0.6.3
+llama4_rope,liger,backward,speed,ms,T,sequence length,2048,0.17812799662351608,0.15929600596427917,0.22060799598693848,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:06,0.6.3
+llama4_rope,liger,backward,speed,ms,T,sequence length,4096,0.1759680062532425,0.15674879848957063,0.21720320284366607,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:06,0.6.3
+llama4_rope,liger,backward,speed,ms,T,sequence length,8192,0.2165440022945404,0.2160000056028366,0.217056006193161,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:06,0.6.3
+llama4_rope,liger,backward,speed,ms,T,sequence length,16384,0.41657599806785583,0.4159039855003357,0.41723520755767823,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:06,0.6.3
+llama4_rope,huggingface,backward,speed,ms,T,sequence length,1024,0.1751679927110672,0.15575680136680603,0.19749760031700136,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:08,0.6.3
+llama4_rope,huggingface,backward,speed,ms,T,sequence length,2048,0.2115200012922287,0.210207998752594,0.21276800334453583,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:08,0.6.3
+llama4_rope,huggingface,backward,speed,ms,T,sequence length,4096,0.3988800048828125,0.3976063966751099,0.39983999729156494,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:08,0.6.3
+llama4_rope,huggingface,backward,speed,ms,T,sequence length,8192,0.755295991897583,0.7539520263671875,0.7564160227775574,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:08,0.6.3
+llama4_rope,huggingface,backward,speed,ms,T,sequence length,16384,1.4835840463638306,1.4825216293334962,1.484928011894226,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:08,0.6.3
+llama4_rope,liger,full,speed,ms,T,sequence length,1024,0.31564798951148987,0.2882175981998444,0.34517760872840886,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:10,0.6.3
+llama4_rope,liger,full,speed,ms,T,sequence length,2048,0.3128319978713989,0.2840575993061066,0.35068159699440005,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:10,0.6.3
+llama4_rope,liger,full,speed,ms,T,sequence length,4096,0.3250559866428375,0.29862399101257325,0.35418241024017333,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:10,0.6.3
+llama4_rope,liger,full,speed,ms,T,sequence length,8192,0.43091198801994324,0.4299199879169464,0.43215999007225037,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:10,0.6.3
+llama4_rope,liger,full,speed,ms,T,sequence length,16384,0.8324480056762695,0.8313279747962952,0.8339200019836426,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:10,0.6.3
+llama4_rope,huggingface,full,speed,ms,T,sequence length,1024,0.2905760109424591,0.2640191912651062,0.3243392050266266,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:13,0.6.3
+llama4_rope,huggingface,full,speed,ms,T,sequence length,2048,0.41596800088882446,0.41514880061149595,0.416787201166153,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:13,0.6.3
+llama4_rope,huggingface,full,speed,ms,T,sequence length,4096,0.7826560139656067,0.7817599773406982,0.783456027507782,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:13,0.6.3
+llama4_rope,huggingface,full,speed,ms,T,sequence length,8192,1.4954240322113037,1.4943679571151733,1.4963200092315674,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:13,0.6.3
+llama4_rope,huggingface,full,speed,ms,T,sequence length,16384,2.9378879070281982,2.936902475357056,2.939033555984497,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:13,0.6.3
+llama4_rope,liger,full,memory,MB,T,sequence length,1024,73.75830078125,73.75830078125,73.75830078125,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:13,0.6.3
+llama4_rope,liger,full,memory,MB,T,sequence length,2048,115.51611328125,115.51611328125,115.51611328125,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:13,0.6.3
+llama4_rope,liger,full,memory,MB,T,sequence length,4096,199.03173828125,199.03173828125,199.03173828125,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:13,0.6.3
+llama4_rope,liger,full,memory,MB,T,sequence length,8192,366.06298828125,366.06298828125,366.06298828125,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:13,0.6.3
+llama4_rope,liger,full,memory,MB,T,sequence length,16384,700.12548828125,700.12548828125,700.12548828125,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:13,0.6.3
+llama4_rope,huggingface,full,memory,MB,T,sequence length,1024,173.00830078125,173.00830078125,173.00830078125,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:13,0.6.3
+llama4_rope,huggingface,full,memory,MB,T,sequence length,2048,314.01611328125,314.01611328125,314.01611328125,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:13,0.6.3
+llama4_rope,huggingface,full,memory,MB,T,sequence length,4096,596.03173828125,596.03173828125,596.03173828125,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:13,0.6.3
+llama4_rope,huggingface,full,memory,MB,T,sequence length,8192,1160.06298828125,1160.06298828125,1160.06298828125,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:13,0.6.3
+llama4_rope,huggingface,full,memory,MB,T,sequence length,16384,2288.12548828125,2288.12548828125,2288.12548828125,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:51:13,0.6.3
+multi_token_attention,liger,forward,speed,ms,L,sequence length,32,0.1279039978981018,0.11032960265874862,0.151500803232193,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:05,0.6.3
+multi_token_attention,liger,forward,speed,ms,L,sequence length,64,0.10089600086212158,0.09438080191612244,0.11935999989509583,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:05,0.6.3
+multi_token_attention,liger,forward,speed,ms,L,sequence length,128,0.11288000270724297,0.09975039958953857,0.12992000281810762,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:05,0.6.3
+multi_token_attention,liger,forward,speed,ms,L,sequence length,256,0.1574079990386963,0.15684479773044585,0.1579391986131668,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:05,0.6.3
+multi_token_attention,liger,forward,speed,ms,L,sequence length,512,0.6880319714546204,0.6865280270576477,0.6894528150558472,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:05,0.6.3
+multi_token_attention,torch,forward,speed,ms,L,sequence length,32,0.15228800475597382,0.14478079974651337,0.1675711989402771,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:06,0.6.3
+multi_token_attention,torch,forward,speed,ms,L,sequence length,64,0.14844800531864166,0.14065279960632324,0.16297600269317628,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:06,0.6.3
+multi_token_attention,torch,forward,speed,ms,L,sequence length,128,0.1483840048313141,0.14049920439720154,0.16295040249824522,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:06,0.6.3
+multi_token_attention,torch,forward,speed,ms,L,sequence length,256,0.14953599870204926,0.14139519929885863,0.1649920016527176,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:06,0.6.3
+multi_token_attention,torch,forward,speed,ms,L,sequence length,512,0.19660800695419312,0.18840960264205933,0.20743680000305176,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:06,0.6.3
+multi_token_attention,liger,full,speed,ms,L,sequence length,32,0.7105599939823151,0.664736020565033,0.7638336062431336,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:15,0.6.3
+multi_token_attention,liger,full,speed,ms,L,sequence length,64,0.7428480088710785,0.7054399967193603,0.7869184017181396,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:15,0.6.3
+multi_token_attention,liger,full,speed,ms,L,sequence length,128,0.7436000108718872,0.709337592124939,0.8501440167427063,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:15,0.6.3
+multi_token_attention,liger,full,speed,ms,L,sequence length,256,0.7447839975357056,0.7065152049064636,0.8237248063087463,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:15,0.6.3
+multi_token_attention,liger,full,speed,ms,L,sequence length,512,1.317728042602539,1.3156479835510253,1.3189695835113526,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:15,0.6.3
+multi_token_attention,torch,full,speed,ms,L,sequence length,32,0.505264014005661,0.48310399055480957,0.5354496002197267,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:16,0.6.3
+multi_token_attention,torch,full,speed,ms,L,sequence length,64,0.559935986995697,0.536575973033905,0.6042879819869995,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:16,0.6.3
+multi_token_attention,torch,full,speed,ms,L,sequence length,128,0.544704020023346,0.5242624163627625,0.5794368147850036,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:16,0.6.3
+multi_token_attention,torch,full,speed,ms,L,sequence length,256,0.5646400153636932,0.541158401966095,0.6005120038986207,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:16,0.6.3
+multi_token_attention,torch,full,speed,ms,L,sequence length,512,0.6822400093078613,0.662559998035431,0.6976063966751098,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:16,0.6.3
+multi_token_attention,liger,backward,speed,ms,L,sequence length,32,0.39500799775123596,0.3642304003238678,0.4427008032798767,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:17,0.6.3
+multi_token_attention,liger,backward,speed,ms,L,sequence length,64,0.4671359956264496,0.42888320684432985,0.5060543775558471,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:17,0.6.3
+multi_token_attention,liger,backward,speed,ms,L,sequence length,128,0.44435200095176697,0.4141248047351837,0.48133120536804197,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:17,0.6.3
+multi_token_attention,liger,backward,speed,ms,L,sequence length,256,0.4273279905319214,0.39633281230926515,0.46148478984832764,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:17,0.6.3
+multi_token_attention,liger,backward,speed,ms,L,sequence length,512,0.6304000020027161,0.628928005695343,0.6317440271377563,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:17,0.6.3
+multi_token_attention,torch,backward,speed,ms,L,sequence length,32,0.20926400274038315,0.192972794175148,0.23721599578857422,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:17,0.6.3
+multi_token_attention,torch,backward,speed,ms,L,sequence length,64,0.2524479925632477,0.2306559979915619,0.3110719919204712,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:17,0.6.3
+multi_token_attention,torch,backward,speed,ms,L,sequence length,128,0.2443840056657791,0.22180479764938354,0.2764095962047577,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:17,0.6.3
+multi_token_attention,torch,backward,speed,ms,L,sequence length,256,0.25513601303100586,0.23326719701290133,0.29877120852470396,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:17,0.6.3
+multi_token_attention,torch,backward,speed,ms,L,sequence length,512,0.3490239977836609,0.34697601199150085,0.35075199604034424,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:17,0.6.3
+multi_token_attention,liger,full,memory,MB,L,sequence length,32,0.19189453125,0.19189453125,0.19189453125,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:17,0.6.3
+multi_token_attention,liger,full,memory,MB,L,sequence length,64,1.568359375,1.568359375,1.568359375,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:17,0.6.3
+multi_token_attention,liger,full,memory,MB,L,sequence length,128,4.146484375,4.146484375,4.146484375,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:17,0.6.3
+multi_token_attention,liger,full,memory,MB,L,sequence length,256,12.912109375,12.912109375,12.912109375,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:17,0.6.3
+multi_token_attention,liger,full,memory,MB,L,sequence length,512,48.912109375,48.912109375,48.912109375,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:17,0.6.3
+multi_token_attention,torch,full,memory,MB,L,sequence length,32,0.17822265625,0.17822265625,0.17822265625,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:17,0.6.3
+multi_token_attention,torch,full,memory,MB,L,sequence length,64,1.51318359375,1.51318359375,1.51318359375,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:17,0.6.3
+multi_token_attention,torch,full,memory,MB,L,sequence length,128,3.92724609375,3.92724609375,3.92724609375,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:17,0.6.3
+multi_token_attention,torch,full,memory,MB,L,sequence length,256,12.03662109375,12.03662109375,12.03662109375,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:17,0.6.3
+multi_token_attention,torch,full,memory,MB,L,sequence length,512,45.41162109375,45.41162109375,45.41162109375,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:52:17,0.6.3
+poly_norm,liger,forward,speed,ms,H,hidden size,1024,0.010015999898314476,0.009696000255644321,0.010239999741315842,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:00,0.6.3
+poly_norm,liger,forward,speed,ms,H,hidden size,2048,0.01398400031030178,0.01369599997997284,0.014303999952971935,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:00,0.6.3
+poly_norm,liger,forward,speed,ms,H,hidden size,4096,0.019967999309301376,0.019807999953627586,0.020191999152302742,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:00,0.6.3
+poly_norm,liger,forward,speed,ms,H,hidden size,8192,0.03500799834728241,0.03471999987959862,0.035392001271247864,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:00,0.6.3
+poly_norm,liger,forward,speed,ms,H,hidden size,16384,0.07206399738788605,0.07135999947786331,0.0727040022611618,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:00,0.6.3
+poly_norm,liger,forward,speed,ms,H,hidden size,32768,0.6563839912414551,0.6548863887786865,0.6577919721603394,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:00,0.6.3
+poly_norm,huggingface,forward,speed,ms,H,hidden size,1024,0.1794240027666092,0.1690559983253479,0.19857919812202457,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:03,0.6.3
+poly_norm,huggingface,forward,speed,ms,H,hidden size,2048,0.27881601452827454,0.2775680124759674,0.2800639867782593,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:03,0.6.3
+poly_norm,huggingface,forward,speed,ms,H,hidden size,4096,0.5701119899749756,0.5689792156219482,0.5712640285491943,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:03,0.6.3
+poly_norm,huggingface,forward,speed,ms,H,hidden size,8192,1.030784010887146,1.0291839838027954,1.0320639610290527,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:03,0.6.3
+poly_norm,huggingface,forward,speed,ms,H,hidden size,16384,1.9452159404754639,1.943769598007202,1.9467008352279662,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:03,0.6.3
+poly_norm,huggingface,forward,speed,ms,H,hidden size,32768,3.762399911880493,3.7608512401580807,3.763967990875244,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:03,0.6.3
+poly_norm,liger,full,speed,ms,H,hidden size,1024,0.24395199865102768,0.2215039998292923,0.2854336082935333,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:08,0.6.3
+poly_norm,liger,full,speed,ms,H,hidden size,2048,0.24879999458789825,0.22444799542427063,0.29727360010147097,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:08,0.6.3
+poly_norm,liger,full,speed,ms,H,hidden size,4096,0.24942400306463242,0.2243904024362564,0.2983424007892609,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:08,0.6.3
+poly_norm,liger,full,speed,ms,H,hidden size,8192,0.2524639964103699,0.22483199834823608,0.30582401156425476,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:08,0.6.3
+poly_norm,liger,full,speed,ms,H,hidden size,16384,0.24326400458812714,0.22351999580860138,0.2918720066547394,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:08,0.6.3
+poly_norm,liger,full,speed,ms,H,hidden size,32768,1.3466880321502686,1.341088056564331,1.3521087646484375,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:08,0.6.3
+poly_norm,huggingface,full,speed,ms,H,hidden size,1024,0.9433919787406921,0.8690175890922547,1.0081727743148803,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:11,0.6.3
+poly_norm,huggingface,full,speed,ms,H,hidden size,2048,1.0038080215454102,1.001030397415161,1.0065856218338012,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:11,0.6.3
+poly_norm,huggingface,full,speed,ms,H,hidden size,4096,1.9705599546432495,1.9677120447158813,1.972991943359375,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:11,0.6.3
+poly_norm,huggingface,full,speed,ms,H,hidden size,8192,3.6088799238204956,3.606169652938843,3.6113215923309325,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:11,0.6.3
+poly_norm,huggingface,full,speed,ms,H,hidden size,16384,6.80131196975708,6.798431873321533,6.804160118103027,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:11,0.6.3
+poly_norm,huggingface,full,speed,ms,H,hidden size,32768,13.160351753234863,13.157075119018554,13.162969779968261,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:11,0.6.3
+poly_norm,liger,backward,speed,ms,H,hidden size,1024,0.1403840035200119,0.12337920069694519,0.17257599532604218,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:14,0.6.3
+poly_norm,liger,backward,speed,ms,H,hidden size,2048,0.14401599764823914,0.12497919797897339,0.18606719672679906,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:14,0.6.3
+poly_norm,liger,backward,speed,ms,H,hidden size,4096,0.1350719928741455,0.12041600048542023,0.16073599457740784,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:14,0.6.3
+poly_norm,liger,backward,speed,ms,H,hidden size,8192,0.14720000326633453,0.12707200646400452,0.1876160055398941,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:14,0.6.3
+poly_norm,liger,backward,speed,ms,H,hidden size,16384,0.14924800395965576,0.12966400384902954,0.17671039700508118,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:14,0.6.3
+poly_norm,liger,backward,speed,ms,H,hidden size,32768,0.692736029624939,0.6875711798667907,0.6985728025436402,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:14,0.6.3
+poly_norm,huggingface,backward,speed,ms,H,hidden size,1024,0.6065599918365479,0.5578304171562195,0.6633535861968993,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:17,0.6.3
+poly_norm,huggingface,backward,speed,ms,H,hidden size,2048,0.7271360158920288,0.725055992603302,0.7291520237922668,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:17,0.6.3
+poly_norm,huggingface,backward,speed,ms,H,hidden size,4096,1.4033279418945312,1.401695966720581,1.4054720401763916,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:17,0.6.3
+poly_norm,huggingface,backward,speed,ms,H,hidden size,8192,2.5793120861053467,2.577120065689087,2.5811967849731445,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:17,0.6.3
+poly_norm,huggingface,backward,speed,ms,H,hidden size,16384,4.857632160186768,4.855654239654541,4.860896110534668,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:17,0.6.3
+poly_norm,huggingface,backward,speed,ms,H,hidden size,32768,9.39731216430664,9.3951997756958,9.400287628173828,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:17,0.6.3
+poly_norm,liger,full,memory,MB,H,hidden size,1024,12.0302734375,12.0302734375,12.0302734375,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:17,0.6.3
+poly_norm,liger,full,memory,MB,H,hidden size,2048,24.0302734375,24.0302734375,24.0302734375,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:17,0.6.3
+poly_norm,liger,full,memory,MB,H,hidden size,4096,48.0302734375,48.0302734375,48.0302734375,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:17,0.6.3
+poly_norm,liger,full,memory,MB,H,hidden size,8192,96.0302734375,96.0302734375,96.0302734375,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:17,0.6.3
+poly_norm,liger,full,memory,MB,H,hidden size,16384,192.0302734375,192.0302734375,192.0302734375,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:17,0.6.3
+poly_norm,liger,full,memory,MB,H,hidden size,32768,384.0302734375,384.0302734375,384.0302734375,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:17,0.6.3
+poly_norm,huggingface,full,memory,MB,H,hidden size,1024,128.0263671875,128.0263671875,128.0263671875,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:18,0.6.3
+poly_norm,huggingface,full,memory,MB,H,hidden size,2048,256.0263671875,256.0263671875,256.0263671875,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:18,0.6.3
+poly_norm,huggingface,full,memory,MB,H,hidden size,4096,512.0263671875,512.0263671875,512.0263671875,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:18,0.6.3
+poly_norm,huggingface,full,memory,MB,H,hidden size,8192,1024.0263671875,1024.0263671875,1024.0263671875,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:18,0.6.3
+poly_norm,huggingface,full,memory,MB,H,hidden size,16384,2048.0263671875,2048.0263671875,2048.0263671875,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:18,0.6.3
+poly_norm,huggingface,full,memory,MB,H,hidden size,32768,4096.0263671875,4096.0263671875,4096.0263671875,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:18,0.6.3
+rms_norm,liger,forward,speed,ms,H,hidden size,1024,0.009600000455975533,0.009503999724984169,0.009824000298976898,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:31,0.6.3
+rms_norm,liger,forward,speed,ms,H,hidden size,2048,0.012543999589979649,0.012223999947309494,0.012799999676644802,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:31,0.6.3
+rms_norm,liger,forward,speed,ms,H,hidden size,4096,0.018592000007629395,0.01817600056529045,0.019071999937295914,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:31,0.6.3
+rms_norm,liger,forward,speed,ms,H,hidden size,8192,0.03094400092959404,0.030751999467611313,0.03171199932694435,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:31,0.6.3
+rms_norm,liger,forward,speed,ms,H,hidden size,16384,0.0634239986538887,0.06284800171852112,0.06412799656391144,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:31,0.6.3
+rms_norm,liger,forward,speed,ms,H,hidden size,32768,0.10947199910879135,0.1090560033917427,0.10995200276374817,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:31,0.6.3
+rms_norm,huggingface,forward,speed,ms,H,hidden size,1024,0.05587200075387955,0.05516799911856651,0.05673599988222122,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:34,0.6.3
+rms_norm,huggingface,forward,speed,ms,H,hidden size,2048,0.08404799923300743,0.08259200304746628,0.08528000116348267,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:34,0.6.3
+rms_norm,huggingface,forward,speed,ms,H,hidden size,4096,0.16988800466060638,0.16838400065898895,0.17132799327373505,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:34,0.6.3
+rms_norm,huggingface,forward,speed,ms,H,hidden size,8192,0.3104639947414398,0.3088639974594116,0.3118399977684021,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:34,0.6.3
+rms_norm,huggingface,forward,speed,ms,H,hidden size,16384,0.5794559717178345,0.5779839754104614,0.5809280276298523,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:34,0.6.3
+rms_norm,huggingface,forward,speed,ms,H,hidden size,32768,1.1146240234375,1.1129920482635498,1.115936040878296,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:34,0.6.3
+rms_norm,liger,full,speed,ms,H,hidden size,1024,0.2579360008239746,0.22725119590759277,0.31284479498863227,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:38,0.6.3
+rms_norm,liger,full,speed,ms,H,hidden size,2048,0.2703680098056793,0.24131199717521667,0.32419198751449585,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:38,0.6.3
+rms_norm,liger,full,speed,ms,H,hidden size,4096,0.2666879892349243,0.23687679469585418,0.3211775898933411,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:38,0.6.3
+rms_norm,liger,full,speed,ms,H,hidden size,8192,0.2645759880542755,0.2311359941959381,0.3213120102882385,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:38,0.6.3
+rms_norm,liger,full,speed,ms,H,hidden size,16384,0.2653439939022064,0.2417471945285797,0.3094399929046631,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:38,0.6.3
+rms_norm,liger,full,speed,ms,H,hidden size,32768,0.9068160057067871,0.9036480188369751,0.9096575856208802,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:38,0.6.3
+rms_norm,huggingface,full,speed,ms,H,hidden size,1024,0.38128000497817993,0.358271986246109,0.4143872022628784,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:42,0.6.3
+rms_norm,huggingface,full,speed,ms,H,hidden size,2048,0.42003199458122253,0.3885183870792389,0.5305343866348267,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:42,0.6.3
+rms_norm,huggingface,full,speed,ms,H,hidden size,4096,0.6144639849662781,0.6132799983024597,0.6155840158462524,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:42,0.6.3
+rms_norm,huggingface,full,speed,ms,H,hidden size,8192,1.1153600215911865,1.1142079830169678,1.1164480447769165,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:42,0.6.3
+rms_norm,huggingface,full,speed,ms,H,hidden size,16384,2.089024066925049,2.087718439102173,2.090502452850342,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:42,0.6.3
+rms_norm,huggingface,full,speed,ms,H,hidden size,32768,4.023392200469971,4.021868705749512,4.024665641784668,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:42,0.6.3
+rms_norm,liger,backward,speed,ms,H,hidden size,1024,0.14601599425077438,0.1309632033109665,0.18222719430923467,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:45,0.6.3
+rms_norm,liger,backward,speed,ms,H,hidden size,2048,0.1454399973154068,0.13067519664764404,0.1895936012268068,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:45,0.6.3
+rms_norm,liger,backward,speed,ms,H,hidden size,4096,0.14372800290584564,0.12796799838542938,0.19597439467906952,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:45,0.6.3
+rms_norm,liger,backward,speed,ms,H,hidden size,8192,0.1398240029811859,0.12512640058994293,0.16921600699424744,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:45,0.6.3
+rms_norm,liger,backward,speed,ms,H,hidden size,16384,0.14633600413799286,0.1331135958433151,0.16828800737857819,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:45,0.6.3
+rms_norm,liger,backward,speed,ms,H,hidden size,32768,0.7982079982757568,0.7953152179718017,0.8011904001235962,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:45,0.6.3
+rms_norm,huggingface,backward,speed,ms,H,hidden size,1024,0.2553919851779938,0.2349119931459427,0.30588799715042114,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:48,0.6.3
+rms_norm,huggingface,backward,speed,ms,H,hidden size,2048,0.2587200105190277,0.2380863964557648,0.30749440789222715,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:48,0.6.3
+rms_norm,huggingface,backward,speed,ms,H,hidden size,4096,0.442656010389328,0.4416320025920868,0.44361600279808044,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:48,0.6.3
+rms_norm,huggingface,backward,speed,ms,H,hidden size,8192,0.8047040104866028,0.8036800026893616,0.8059008002281189,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:48,0.6.3
+rms_norm,huggingface,backward,speed,ms,H,hidden size,16384,1.5134400129318237,1.5122239589691162,1.5146559476852417,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:48,0.6.3
+rms_norm,huggingface,backward,speed,ms,H,hidden size,32768,2.9154880046844482,2.9143359661102295,2.916972780227661,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:48,0.6.3
+rms_norm,liger,full,memory,MB,H,hidden size,1024,12.5390625,12.5390625,12.5390625,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:48,0.6.3
+rms_norm,liger,full,memory,MB,H,hidden size,2048,25.0703125,25.0703125,25.0703125,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:48,0.6.3
+rms_norm,liger,full,memory,MB,H,hidden size,4096,50.1328125,50.1328125,50.1328125,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:48,0.6.3
+rms_norm,liger,full,memory,MB,H,hidden size,8192,100.2578125,100.2578125,100.2578125,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:48,0.6.3
+rms_norm,liger,full,memory,MB,H,hidden size,16384,200.5078125,200.5078125,200.5078125,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:48,0.6.3
+rms_norm,liger,full,memory,MB,H,hidden size,32768,401.0078125,401.0078125,401.0078125,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:48,0.6.3
+rms_norm,huggingface,full,memory,MB,H,hidden size,1024,80.01953125,80.01953125,80.01953125,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:48,0.6.3
+rms_norm,huggingface,full,memory,MB,H,hidden size,2048,160.03125,160.03125,160.03125,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:48,0.6.3
+rms_norm,huggingface,full,memory,MB,H,hidden size,4096,320.0546875,320.0546875,320.0546875,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:48,0.6.3
+rms_norm,huggingface,full,memory,MB,H,hidden size,8192,640.1015625,640.1015625,640.1015625,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:48,0.6.3
+rms_norm,huggingface,full,memory,MB,H,hidden size,16384,1280.1953125,1280.1953125,1280.1953125,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:48,0.6.3
+rms_norm,huggingface,full,memory,MB,H,hidden size,32768,2560.3828125,2560.3828125,2560.3828125,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:48,0.6.3
+rope,liger,forward,speed,ms,H,hidden size,512,0.007872000336647034,0.007679999805986881,0.0080960001796484,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:57,0.6.3
+rope,liger,forward,speed,ms,H,hidden size,2048,0.013407999649643898,0.01321600005030632,0.013632000423967838,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:57,0.6.3
+rope,liger,forward,speed,ms,H,hidden size,8192,0.03411199897527695,0.03379200026392937,0.03484800085425377,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:57,0.6.3
+rope,huggingface,forward,speed,ms,H,hidden size,512,0.06022400036454201,0.059487998485565186,0.06140799820423126,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:58,0.6.3
+rope,huggingface,forward,speed,ms,H,hidden size,2048,0.10646399855613708,0.10585600137710571,0.10732799768447876,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:58,0.6.3
+rope,huggingface,forward,speed,ms,H,hidden size,8192,0.32707199454307556,0.32607999444007874,0.32837761044502256,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:53:58,0.6.3
+rope,liger,backward,speed,ms,H,hidden size,512,0.09654399752616882,0.08348800241947174,0.11727999895811081,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:00,0.6.3
+rope,liger,backward,speed,ms,H,hidden size,2048,0.10116800293326378,0.08899839967489243,0.1208832010626793,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:00,0.6.3
+rope,liger,backward,speed,ms,H,hidden size,8192,0.09676799923181534,0.08451200276613235,0.11366400122642517,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:00,0.6.3
+rope,huggingface,backward,speed,ms,H,hidden size,512,0.2544640004634857,0.23438079953193663,0.28575998544692993,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:01,0.6.3
+rope,huggingface,backward,speed,ms,H,hidden size,2048,0.24544000625610352,0.22096640467643736,0.27486719489097594,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:01,0.6.3
+rope,huggingface,backward,speed,ms,H,hidden size,8192,0.38979199528694153,0.3887999951839447,0.39129599928855896,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:01,0.6.3
+rope,liger,full,speed,ms,H,hidden size,512,0.174112007021904,0.15600000321865082,0.21160959601402285,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:02,0.6.3
+rope,liger,full,speed,ms,H,hidden size,2048,0.17632000148296356,0.1561344027519226,0.20776960253715518,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:02,0.6.3
+rope,liger,full,speed,ms,H,hidden size,8192,0.17030400037765503,0.15190400183200836,0.20057600736618042,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:02,0.6.3
+rope,huggingface,full,speed,ms,H,hidden size,512,0.3808639943599701,0.35604481101036073,0.41052801012992857,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:04,0.6.3
+rope,huggingface,full,speed,ms,H,hidden size,2048,0.4156000018119812,0.39294078946113586,0.4493119895458222,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:04,0.6.3
+rope,huggingface,full,speed,ms,H,hidden size,8192,0.7144320011138916,0.7130879759788513,0.7158399820327759,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:04,0.6.3
+rope,liger,full,memory,MB,H,hidden size,512,37.14111328125,37.14111328125,37.14111328125,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:04,0.6.3
+rope,liger,full,memory,MB,H,hidden size,2048,52.51611328125,52.51611328125,52.51611328125,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:04,0.6.3
+rope,liger,full,memory,MB,H,hidden size,8192,114.01611328125,114.01611328125,114.01611328125,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:04,0.6.3
+rope,huggingface,full,memory,MB,H,hidden size,512,46.14111328125,46.14111328125,46.14111328125,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:04,0.6.3
+rope,huggingface,full,memory,MB,H,hidden size,2048,88.51611328125,88.51611328125,88.51611328125,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:04,0.6.3
+rope,huggingface,full,memory,MB,H,hidden size,8192,258.01611328125,258.01611328125,258.01611328125,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:04,0.6.3
+rope,liger,forward,speed,ms,T,sequence length,1024,0.01945599913597107,0.019168000668287277,0.019936000928282738,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:06,0.6.3
+rope,liger,forward,speed,ms,T,sequence length,2048,0.034143999218940735,0.03385600075125694,0.034912001341581345,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:06,0.6.3
+rope,liger,forward,speed,ms,T,sequence length,4096,0.06297600269317627,0.062144000083208084,0.0634239986538887,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:06,0.6.3
+rope,liger,forward,speed,ms,T,sequence length,8192,0.11900799721479416,0.11868800222873688,0.1194240003824234,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:06,0.6.3
+rope,liger,forward,speed,ms,T,sequence length,16384,0.23164799809455872,0.23126399517059326,0.23231999576091766,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:06,0.6.3
+rope,huggingface,forward,speed,ms,T,sequence length,1024,0.17961600422859192,0.17865599691867828,0.18086400628089905,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:08,0.6.3
+rope,huggingface,forward,speed,ms,T,sequence length,2048,0.3271999955177307,0.3261759877204895,0.3284800052642822,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:08,0.6.3
+rope,huggingface,forward,speed,ms,T,sequence length,4096,0.6134080290794373,0.61244797706604,0.6147519946098328,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:08,0.6.3
+rope,huggingface,forward,speed,ms,T,sequence length,8192,1.1828160285949707,1.1819839477539062,1.1839359998703003,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:08,0.6.3
+rope,huggingface,forward,speed,ms,T,sequence length,16384,2.313839912414551,2.3117759227752686,2.315808057785034,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:08,0.6.3
+rope,liger,backward,speed,ms,T,sequence length,1024,0.1064319983124733,0.09433600306510925,0.12255360186100006,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:10,0.6.3
+rope,liger,backward,speed,ms,T,sequence length,2048,0.10513600334525108,0.09228800237178802,0.1260479986667633,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:10,0.6.3
+rope,liger,backward,speed,ms,T,sequence length,4096,0.10488000139594078,0.09427200257778168,0.12377599775791169,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:10,0.6.3
+rope,liger,backward,speed,ms,T,sequence length,8192,0.11929599940776825,0.11881600320339203,0.12115839868783951,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:10,0.6.3
+rope,liger,backward,speed,ms,T,sequence length,16384,0.2314240038394928,0.23100159466266632,0.2322559952735901,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:10,0.6.3
+rope,huggingface,backward,speed,ms,T,sequence length,1024,0.28035199642181396,0.264384001493454,0.3178048133850099,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:12,0.6.3
+rope,huggingface,backward,speed,ms,T,sequence length,2048,0.3927359879016876,0.38981119990348817,0.3981119990348816,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:12,0.6.3
+rope,huggingface,backward,speed,ms,T,sequence length,4096,0.7331519722938538,0.7299200296401978,0.7362560033798218,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:12,0.6.3
+rope,huggingface,backward,speed,ms,T,sequence length,8192,1.4029439687728882,1.398956823348999,1.4062976121902466,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:12,0.6.3
+rope,huggingface,backward,speed,ms,T,sequence length,16384,2.725503921508789,2.721280097961426,2.7300479412078857,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:12,0.6.3
+rope,liger,full,speed,ms,T,sequence length,1024,0.19353599846363068,0.17635199427604675,0.22579840421676636,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:14,0.6.3
+rope,liger,full,speed,ms,T,sequence length,2048,0.1841759979724884,0.16347519755363465,0.2159936010837555,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:14,0.6.3
+rope,liger,full,speed,ms,T,sequence length,4096,0.16156800091266632,0.14339199662208557,0.18985599279403687,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:14,0.6.3
+rope,liger,full,speed,ms,T,sequence length,8192,0.23603199422359467,0.2351360023021698,0.2375359982252121,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:14,0.6.3
+rope,liger,full,speed,ms,T,sequence length,16384,0.46060800552368164,0.45983999967575073,0.46134400367736816,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:14,0.6.3
+rope,huggingface,full,speed,ms,T,sequence length,1024,0.3892480134963989,0.38630399107933044,0.39232000708580017,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:16,0.6.3
+rope,huggingface,full,speed,ms,T,sequence length,2048,0.7126399874687195,0.711359977722168,0.7139520049095154,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:16,0.6.3
+rope,huggingface,full,speed,ms,T,sequence length,4096,1.3367680311203003,1.3354560136795044,1.3379199504852295,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:16,0.6.3
+rope,huggingface,full,speed,ms,T,sequence length,8192,2.5782560110092163,2.5766015529632567,2.580390453338623,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:16,0.6.3
+rope,huggingface,full,speed,ms,T,sequence length,16384,5.030752182006836,5.028883266448974,5.034220886230469,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:16,0.6.3
+rope,liger,full,memory,MB,T,sequence length,1024,73.00830078125,73.00830078125,73.00830078125,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:16,0.6.3
+rope,liger,full,memory,MB,T,sequence length,2048,114.01611328125,114.01611328125,114.01611328125,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:16,0.6.3
+rope,liger,full,memory,MB,T,sequence length,4096,196.03173828125,196.03173828125,196.03173828125,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:16,0.6.3
+rope,liger,full,memory,MB,T,sequence length,8192,360.06298828125,360.06298828125,360.06298828125,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:16,0.6.3
+rope,liger,full,memory,MB,T,sequence length,16384,688.12548828125,688.12548828125,688.12548828125,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:16,0.6.3
+rope,huggingface,full,memory,MB,T,sequence length,1024,145.00830078125,145.00830078125,145.00830078125,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:16,0.6.3
+rope,huggingface,full,memory,MB,T,sequence length,2048,258.01611328125,258.01611328125,258.01611328125,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:16,0.6.3
+rope,huggingface,full,memory,MB,T,sequence length,4096,484.03173828125,484.03173828125,484.03173828125,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:16,0.6.3
+rope,huggingface,full,memory,MB,T,sequence length,8192,936.06298828125,936.06298828125,936.06298828125,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:16,0.6.3
+rope,huggingface,full,memory,MB,T,sequence length,16384,1840.12548828125,1840.12548828125,1840.12548828125,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-11-21 00:54:16,0.6.3
+fused_linear_simpo_loss,liger,forward,speed,ms,B,B,2,12.003695964813232,11.847628593444824,12.180000114440919,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:55:04,0.6.3
+fused_linear_simpo_loss,liger,forward,speed,ms,B,B,4,23.96294403076172,23.925945281982422,24.021177673339842,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:55:04,0.6.3
+fused_linear_simpo_loss,liger,forward,speed,ms,B,B,8,47.31955146789551,46.90997772216797,47.72912521362305,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:55:04,0.6.3
+fused_linear_simpo_loss,liger,forward,speed,ms,B,B,16,95.21923065185547,95.21923065185547,95.21923065185547,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:55:04,0.6.3
+fused_linear_simpo_loss,huggingface,forward,speed,ms,B,B,2,6.029776096343994,6.024288177490234,6.0386881828308105,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:55:27,0.6.3
+fused_linear_simpo_loss,huggingface,forward,speed,ms,B,B,4,11.832304000854492,11.819193267822264,11.842130851745605,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:55:27,0.6.3
+fused_linear_simpo_loss,huggingface,forward,speed,ms,B,B,8,23.21110439300537,23.169638442993165,23.26049346923828,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:55:27,0.6.3
+fused_linear_simpo_loss,huggingface,forward,speed,ms,B,B,16,46.45662307739258,46.425682830810544,46.48756332397461,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:55:27,0.6.3
+fused_linear_simpo_loss,liger,full,speed,ms,B,B,2,13.974256038665771,13.8571195602417,14.525216102600098,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:55:52,0.6.3
+fused_linear_simpo_loss,liger,full,speed,ms,B,B,4,26.412832260131836,25.958330154418945,26.457030487060546,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:55:52,0.6.3
+fused_linear_simpo_loss,liger,full,speed,ms,B,B,8,49.51947212219238,49.465395355224615,49.57354888916016,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:55:52,0.6.3
+fused_linear_simpo_loss,liger,full,speed,ms,B,B,16,97.30252838134766,97.30252838134766,97.30252838134766,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:55:52,0.6.3
+fused_linear_simpo_loss,huggingface,full,speed,ms,B,B,2,17.19308853149414,17.171532058715822,17.795513534545897,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:56:19,0.6.3
+fused_linear_simpo_loss,huggingface,full,speed,ms,B,B,4,33.172319412231445,33.00558700561524,33.33905181884766,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:56:19,0.6.3
+fused_linear_simpo_loss,huggingface,full,speed,ms,B,B,8,64.28422546386719,64.28422546386719,64.28422546386719,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:56:19,0.6.3
+fused_linear_simpo_loss,huggingface,full,speed,ms,B,B,16,130.30706787109375,130.30706787109375,130.30706787109375,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:56:19,0.6.3
+fused_linear_simpo_loss,liger,backward,speed,ms,B,B,2,1.0924479961395264,1.0878527641296387,1.1023871898651123,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:56:44,0.6.3
+fused_linear_simpo_loss,liger,backward,speed,ms,B,B,4,1.103551983833313,1.1002624034881592,1.1129087924957275,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:56:44,0.6.3
+fused_linear_simpo_loss,liger,backward,speed,ms,B,B,8,1.12608003616333,1.122451162338257,1.1380864143371583,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:56:44,0.6.3
+fused_linear_simpo_loss,liger,backward,speed,ms,B,B,16,1.167423963546753,1.1649919748306274,1.1724799871444702,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:56:44,0.6.3
+fused_linear_simpo_loss,huggingface,backward,speed,ms,B,B,2,10.965231895446777,10.959897422790528,10.971622276306153,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:57:10,0.6.3
+fused_linear_simpo_loss,huggingface,backward,speed,ms,B,B,4,20.84569549560547,20.833459091186523,20.879269790649417,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:57:10,0.6.3
+fused_linear_simpo_loss,huggingface,backward,speed,ms,B,B,8,41.04355239868164,40.91268463134766,41.17442016601562,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:57:10,0.6.3
+fused_linear_simpo_loss,huggingface,backward,speed,ms,B,B,16,81.1418228149414,81.1418228149414,81.1418228149414,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:57:10,0.6.3
+fused_linear_simpo_loss,liger,full,memory,MB,B,B,2,6142.03466796875,6142.03466796875,6142.03466796875,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:57:35,0.6.3
+fused_linear_simpo_loss,liger,full,memory,MB,B,B,4,6190.0517578125,6190.0517578125,6190.0517578125,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:57:35,0.6.3
+fused_linear_simpo_loss,liger,full,memory,MB,B,B,8,6286.0859375,6286.0859375,6286.0859375,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:57:35,0.6.3
+fused_linear_simpo_loss,liger,full,memory,MB,B,B,16,6478.154296875,6478.154296875,6478.154296875,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:57:35,0.6.3
+fused_linear_simpo_loss,huggingface,full,memory,MB,B,B,2,6610.017578125,6610.017578125,6610.017578125,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:02,0.6.3
+fused_linear_simpo_loss,huggingface,full,memory,MB,B,B,4,10148.033203125,10148.033203125,10148.033203125,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:02,0.6.3
+fused_linear_simpo_loss,huggingface,full,memory,MB,B,B,8,17226.064453125,17226.064453125,17226.064453125,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:02,0.6.3
+fused_linear_simpo_loss,huggingface,full,memory,MB,B,B,16,31382.126953125,31382.126953125,31382.126953125,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:02,0.6.3
+softmax,liger,forward,speed,ms,N,hidden size,128,0.0072639998979866505,0.007135999854654074,0.007519999984651804,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:10,0.6.3
+softmax,liger,forward,speed,ms,N,hidden size,256,0.007552000228315592,0.007424000184983015,0.007679999805986881,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:10,0.6.3
+softmax,liger,forward,speed,ms,N,hidden size,512,0.008704000152647495,0.00848000030964613,0.008895999751985073,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:10,0.6.3
+softmax,liger,forward,speed,ms,N,hidden size,1024,0.011776000261306763,0.01158399973064661,0.01196799986064434,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:10,0.6.3
+softmax,liger,forward,speed,ms,N,hidden size,2048,0.017664000391960144,0.017472000792622566,0.017952000722289085,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:10,0.6.3
+softmax,liger,forward,speed,ms,N,hidden size,4096,0.02828799933195114,0.028031999245285988,0.028736000880599022,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:10,0.6.3
+softmax,torch,forward,speed,ms,N,hidden size,128,0.007135999854654074,0.007040000054985285,0.0072639998979866505,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:14,0.6.3
+softmax,torch,forward,speed,ms,N,hidden size,256,0.007968000136315823,0.00774399982765317,0.0081599997356534,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:14,0.6.3
+softmax,torch,forward,speed,ms,N,hidden size,512,0.009631999768316746,0.009472000412642956,0.009759999811649323,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:14,0.6.3
+softmax,torch,forward,speed,ms,N,hidden size,1024,0.012992000207304955,0.012799999676644802,0.01321600005030632,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:14,0.6.3
+softmax,torch,forward,speed,ms,N,hidden size,2048,0.0197759997099638,0.019519999623298645,0.019999999552965164,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:14,0.6.3
+softmax,torch,forward,speed,ms,N,hidden size,4096,0.036896001547575,0.03667199984192848,0.03731200098991394,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:14,0.6.3
+softmax,liger,full,speed,ms,N,hidden size,128,0.19420799612998962,0.15596800148487092,0.2922432065010071,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:17,0.6.3
+softmax,liger,full,speed,ms,N,hidden size,256,0.15187199413776398,0.13964800536632538,0.18325120210647583,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:17,0.6.3
+softmax,liger,full,speed,ms,N,hidden size,512,0.15094399452209473,0.1392384022474289,0.1712512016296387,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:17,0.6.3
+softmax,liger,full,speed,ms,N,hidden size,1024,0.1520320028066635,0.1401599943637848,0.17604480683803558,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:17,0.6.3
+softmax,liger,full,speed,ms,N,hidden size,2048,0.15343999862670898,0.14073599874973297,0.19034240245819092,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:17,0.6.3
+softmax,liger,full,speed,ms,N,hidden size,4096,0.1520960032939911,0.1390463948249817,0.18262400329113015,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:17,0.6.3
+softmax,torch,full,speed,ms,N,hidden size,128,0.05820799991488457,0.05145600065588951,0.07065600156784058,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:20,0.6.3
+softmax,torch,full,speed,ms,N,hidden size,256,0.05692800134420395,0.049984000623226166,0.06691200286149979,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:20,0.6.3
+softmax,torch,full,speed,ms,N,hidden size,512,0.0568000003695488,0.0492543987929821,0.0689919963479042,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:20,0.6.3
+softmax,torch,full,speed,ms,N,hidden size,1024,0.05584000051021576,0.04914560168981552,0.06656640022993088,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:20,0.6.3
+softmax,torch,full,speed,ms,N,hidden size,2048,0.05867199972271919,0.05302400141954422,0.06972160190343857,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:20,0.6.3
+softmax,torch,full,speed,ms,N,hidden size,4096,0.10915199667215347,0.10864000022411346,0.11075200140476227,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:20,0.6.3
+softmax,liger,backward,speed,ms,N,hidden size,128,0.0756480023264885,0.066880002617836,0.09059199690818787,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:24,0.6.3
+softmax,liger,backward,speed,ms,N,hidden size,256,0.0751039981842041,0.06675200164318085,0.08857599645853043,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:24,0.6.3
+softmax,liger,backward,speed,ms,N,hidden size,512,0.07523199915885925,0.06694400310516357,0.08832640200853348,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:24,0.6.3
+softmax,liger,backward,speed,ms,N,hidden size,1024,0.0735040009021759,0.06524799764156342,0.08691839873790742,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:24,0.6.3
+softmax,liger,backward,speed,ms,N,hidden size,2048,0.07448000088334084,0.06617599725723267,0.08718080073595047,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:24,0.6.3
+softmax,liger,backward,speed,ms,N,hidden size,4096,0.07622399926185608,0.0666240006685257,0.09132160246372223,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:24,0.6.3
+softmax,torch,backward,speed,ms,N,hidden size,128,0.03580800071358681,0.029791999608278275,0.04598399996757507,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:27,0.6.3
+softmax,torch,backward,speed,ms,N,hidden size,256,0.036479998379945755,0.030047999694943428,0.046112000942230225,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:27,0.6.3
+softmax,torch,backward,speed,ms,N,hidden size,512,0.0363520011305809,0.03049599938094616,0.04575999826192856,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:27,0.6.3
+softmax,torch,backward,speed,ms,N,hidden size,1024,0.0363520011305809,0.03062400035560131,0.04499839991331102,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:27,0.6.3
+softmax,torch,backward,speed,ms,N,hidden size,2048,0.04095999896526337,0.03993599861860275,0.04474240094423294,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:27,0.6.3
+softmax,torch,backward,speed,ms,N,hidden size,4096,0.07715199887752533,0.07638400048017502,0.078015998005867,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:27,0.6.3
+softmax,liger,forward,speed,ms,N,hidden size,128,0.007199999876320362,0.006976000033318996,0.007391999941319227,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:30,0.6.3
+softmax,liger,forward,speed,ms,N,hidden size,256,0.007519999984651804,0.007327999919652939,0.011168000288307667,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:30,0.6.3
+softmax,liger,forward,speed,ms,N,hidden size,512,0.007872000336647034,0.007648000027984381,0.012959999963641167,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:30,0.6.3
+softmax,liger,forward,speed,ms,N,hidden size,1024,0.008991999551653862,0.008767999708652496,0.012703999876976013,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:30,0.6.3
+softmax,liger,forward,speed,ms,N,hidden size,2048,0.012160000391304493,0.011744000017642975,0.01587199978530407,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:30,0.6.3
+softmax,liger,forward,speed,ms,N,hidden size,4096,0.01849599927663803,0.017696000635623932,0.020640000700950623,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:30,0.6.3
+softmax,torch,forward,speed,ms,N,hidden size,128,0.007519999984651804,0.0071680000983178616,0.011296000331640244,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:33,0.6.3
+softmax,torch,forward,speed,ms,N,hidden size,256,0.007712000049650669,0.007391999941319227,0.011487999930977821,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:33,0.6.3
+softmax,torch,forward,speed,ms,N,hidden size,512,0.009216000325977802,0.009056000038981438,0.009440000168979168,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:33,0.6.3
+softmax,torch,forward,speed,ms,N,hidden size,1024,0.01228800043463707,0.012128000147640705,0.01244799979031086,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:33,0.6.3
+softmax,torch,forward,speed,ms,N,hidden size,2048,0.018559999763965607,0.018400000408291817,0.018751999363303185,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:33,0.6.3
+softmax,torch,forward,speed,ms,N,hidden size,4096,0.0331839993596077,0.03302399814128876,0.033440001308918,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:33,0.6.3
+softmax,liger,full,speed,ms,N,hidden size,128,0.1568159982562065,0.1430400013923645,0.18533120155334473,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:36,0.6.3
+softmax,liger,full,speed,ms,N,hidden size,256,0.15689599514007568,0.14361600577831268,0.19382399320602417,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:36,0.6.3
+softmax,liger,full,speed,ms,N,hidden size,512,0.16102400422096252,0.1459520012140274,0.21066880226135265,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:36,0.6.3
+softmax,liger,full,speed,ms,N,hidden size,1024,0.15863999724388123,0.1445312023162842,0.19523200392723083,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:36,0.6.3
+softmax,liger,full,speed,ms,N,hidden size,2048,0.16022400557994843,0.1451839953660965,0.2002048045396806,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:36,0.6.3
+softmax,liger,full,speed,ms,N,hidden size,4096,0.16078399866819382,0.14735999703407288,0.19578240215778353,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:36,0.6.3
+softmax,torch,full,speed,ms,N,hidden size,128,0.061535999178886414,0.05302400141954422,0.07544320076704027,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:39,0.6.3
+softmax,torch,full,speed,ms,N,hidden size,256,0.061055999249219894,0.05215999856591225,0.07372800260782242,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:39,0.6.3
+softmax,torch,full,speed,ms,N,hidden size,512,0.06115199998021126,0.05283199995756149,0.0760127976536751,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:39,0.6.3
+softmax,torch,full,speed,ms,N,hidden size,1024,0.05951999872922897,0.051392000168561935,0.07211519777774811,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:39,0.6.3
+softmax,torch,full,speed,ms,N,hidden size,2048,0.058848001062870026,0.05152000114321709,0.0703359991312027,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:39,0.6.3
+softmax,torch,full,speed,ms,N,hidden size,4096,0.0655680000782013,0.06499200314283371,0.06689920276403427,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:39,0.6.3
+softmax,liger,backward,speed,ms,N,hidden size,128,0.07891199737787247,0.06867840141057968,0.09446399658918381,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:42,0.6.3
+softmax,liger,backward,speed,ms,N,hidden size,256,0.07999999821186066,0.06857600063085556,0.09609600156545639,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:42,0.6.3
+softmax,liger,backward,speed,ms,N,hidden size,512,0.07574400305747986,0.06592000275850296,0.09024640172719958,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:42,0.6.3
+softmax,liger,backward,speed,ms,N,hidden size,1024,0.07769600301980972,0.0689919963479042,0.09110400080680847,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:42,0.6.3
+softmax,liger,backward,speed,ms,N,hidden size,2048,0.07635200023651123,0.06800640225410462,0.0902272015810013,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:42,0.6.3
+softmax,liger,backward,speed,ms,N,hidden size,4096,0.07523199915885925,0.06758400052785873,0.08611200004816055,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:42,0.6.3
+softmax,torch,backward,speed,ms,N,hidden size,128,0.034143999218940735,0.02816000021994114,0.042073599994182594,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,torch,backward,speed,ms,N,hidden size,256,0.04179200157523155,0.03230080008506775,0.05552000179886818,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,torch,backward,speed,ms,N,hidden size,512,0.044576000422239304,0.034912001341581345,0.05824640020728113,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,torch,backward,speed,ms,N,hidden size,1024,0.03951999917626381,0.03187200054526329,0.05532800033688545,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,torch,backward,speed,ms,N,hidden size,2048,0.039264000952243805,0.03248000144958496,0.05004800111055374,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,torch,backward,speed,ms,N,hidden size,4096,0.04095999896526337,0.040063999593257904,0.04233599826693535,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,liger,full,memory,MB,N,hidden size,128,0.00244140625,0.00244140625,0.00244140625,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,liger,full,memory,MB,N,hidden size,256,0.0048828125,0.0048828125,0.0048828125,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,liger,full,memory,MB,N,hidden size,512,0.009765625,0.009765625,0.009765625,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,liger,full,memory,MB,N,hidden size,1024,0.01953125,0.01953125,0.01953125,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,liger,full,memory,MB,N,hidden size,2048,0.0390625,0.0390625,0.0390625,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,liger,full,memory,MB,N,hidden size,4096,0.078125,0.078125,0.078125,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,torch,full,memory,MB,N,hidden size,128,0.0029296875,0.0029296875,0.0029296875,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,torch,full,memory,MB,N,hidden size,256,0.005859375,0.005859375,0.005859375,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,torch,full,memory,MB,N,hidden size,512,0.01171875,0.01171875,0.01171875,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,torch,full,memory,MB,N,hidden size,1024,0.0234375,0.0234375,0.0234375,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,torch,full,memory,MB,N,hidden size,2048,0.046875,0.046875,0.046875,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,torch,full,memory,MB,N,hidden size,4096,0.09375,0.09375,0.09375,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,liger,full,memory,MB,N,hidden size,128,0.00244140625,0.00244140625,0.00244140625,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,liger,full,memory,MB,N,hidden size,256,0.00244140625,0.00244140625,0.00244140625,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,liger,full,memory,MB,N,hidden size,512,0.0048828125,0.0048828125,0.0048828125,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,liger,full,memory,MB,N,hidden size,1024,0.009765625,0.009765625,0.009765625,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,liger,full,memory,MB,N,hidden size,2048,0.01953125,0.01953125,0.01953125,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,liger,full,memory,MB,N,hidden size,4096,0.0390625,0.0390625,0.0390625,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,torch,full,memory,MB,N,hidden size,128,0.0029296875,0.0029296875,0.0029296875,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,torch,full,memory,MB,N,hidden size,256,0.0029296875,0.0029296875,0.0029296875,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,torch,full,memory,MB,N,hidden size,512,0.005859375,0.005859375,0.005859375,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,torch,full,memory,MB,N,hidden size,1024,0.01171875,0.01171875,0.01171875,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,torch,full,memory,MB,N,hidden size,2048,0.0234375,0.0234375,0.0234375,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+softmax,torch,full,memory,MB,N,hidden size,4096,0.046875,0.046875,0.046875,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 00:58:46,0.6.3
+sparse_multi_token_attention,liger,forward,speed,ms,L,sequence length,32,0.22191999852657318,0.2075455993413925,0.25233280658721924,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:36,0.6.3
+sparse_multi_token_attention,liger,forward,speed,ms,L,sequence length,64,0.2069759964942932,0.19468800723552704,0.23751040101051332,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:36,0.6.3
+sparse_multi_token_attention,liger,forward,speed,ms,L,sequence length,128,0.19648000597953796,0.18675200641155243,0.22483199834823608,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:36,0.6.3
+sparse_multi_token_attention,liger,forward,speed,ms,L,sequence length,256,0.2592960000038147,0.2584959864616394,0.26015999913215637,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:36,0.6.3
+sparse_multi_token_attention,liger,forward,speed,ms,L,sequence length,512,0.7648640275001526,0.7622079849243164,0.7678016185760498,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:36,0.6.3
+sparse_multi_token_attention,torch,forward,speed,ms,L,sequence length,32,0.4755680114030838,0.4485183894634247,0.49635199308395384,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:37,0.6.3
+sparse_multi_token_attention,torch,forward,speed,ms,L,sequence length,64,0.45395201444625854,0.43525758981704715,0.47266559600830077,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:37,0.6.3
+sparse_multi_token_attention,torch,forward,speed,ms,L,sequence length,128,0.47152000665664673,0.4511424064636231,0.493939197063446,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:37,0.6.3
+sparse_multi_token_attention,torch,forward,speed,ms,L,sequence length,256,0.4620800018310547,0.4397248089313507,0.4864512085914612,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:37,0.6.3
+sparse_multi_token_attention,torch,forward,speed,ms,L,sequence length,512,0.498879998922348,0.48391039967536925,0.5214656114578247,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:37,0.6.3
+sparse_multi_token_attention,liger,full,speed,ms,L,sequence length,32,0.8815360069274902,0.8398720145225524,0.9908543825149536,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:50,0.6.3
+sparse_multi_token_attention,liger,full,speed,ms,L,sequence length,64,0.9259520173072815,0.8837823867797852,1.0398912191390992,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:50,0.6.3
+sparse_multi_token_attention,liger,full,speed,ms,L,sequence length,128,0.9232320189476013,0.8627583980560303,1.024832010269165,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:50,0.6.3
+sparse_multi_token_attention,liger,full,speed,ms,L,sequence length,256,0.8544640243053436,0.8136640191078186,0.9102399945259094,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:50,0.6.3
+sparse_multi_token_attention,liger,full,speed,ms,L,sequence length,512,1.5320639610290527,1.5224640369415283,1.5429439544677734,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:50,0.6.3
+sparse_multi_token_attention,torch,full,speed,ms,L,sequence length,32,1.0853439569473267,1.0748160123825072,1.1139967679977418,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:50,0.6.3
+sparse_multi_token_attention,torch,full,speed,ms,L,sequence length,64,1.1079679727554321,1.084671974182129,1.1632447719573975,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:50,0.6.3
+sparse_multi_token_attention,torch,full,speed,ms,L,sequence length,128,1.1054400205612183,1.0861824035644532,1.1343103647232056,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:50,0.6.3
+sparse_multi_token_attention,torch,full,speed,ms,L,sequence length,256,1.1397759914398193,1.1026432275772096,1.1965184211730957,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:50,0.6.3
+sparse_multi_token_attention,torch,full,speed,ms,L,sequence length,512,1.1619199514389038,1.1299967765808105,1.2072256088256836,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:50,0.6.3
+sparse_multi_token_attention,liger,backward,speed,ms,L,sequence length,32,0.3551360070705414,0.328358393907547,0.39107200503349304,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:51,0.6.3
+sparse_multi_token_attention,liger,backward,speed,ms,L,sequence length,64,0.38967999815940857,0.3677055954933166,0.4178432106971741,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:51,0.6.3
+sparse_multi_token_attention,liger,backward,speed,ms,L,sequence length,128,0.3975679874420166,0.3716735899448395,0.428767991065979,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:51,0.6.3
+sparse_multi_token_attention,liger,backward,speed,ms,L,sequence length,256,0.37648001313209534,0.35172479748725893,0.40245760083198545,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:51,0.6.3
+sparse_multi_token_attention,liger,backward,speed,ms,L,sequence length,512,0.7385280132293701,0.7324991941452026,0.7427008032798768,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:51,0.6.3
+sparse_multi_token_attention,torch,backward,speed,ms,L,sequence length,32,0.3243360072374344,0.3070400059223175,0.3491840064525604,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:52,0.6.3
+sparse_multi_token_attention,torch,backward,speed,ms,L,sequence length,64,0.33961600065231323,0.32350720167160035,0.3563136041164398,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:52,0.6.3
+sparse_multi_token_attention,torch,backward,speed,ms,L,sequence length,128,0.3441600054502487,0.3287679970264435,0.3863296031951905,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:52,0.6.3
+sparse_multi_token_attention,torch,backward,speed,ms,L,sequence length,256,0.32688000798225403,0.3078784108161926,0.3488192021846771,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:52,0.6.3
+sparse_multi_token_attention,torch,backward,speed,ms,L,sequence length,512,0.4128960072994232,0.4107840061187744,0.4148672044277191,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:52,0.6.3
+sparse_multi_token_attention,liger,full,memory,MB,L,sequence length,32,0.318359375,0.318359375,0.318359375,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:52,0.6.3
+sparse_multi_token_attention,liger,full,memory,MB,L,sequence length,64,1.3818359375,1.3818359375,1.3818359375,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:52,0.6.3
+sparse_multi_token_attention,liger,full,memory,MB,L,sequence length,128,5.44091796875,5.44091796875,5.44091796875,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:52,0.6.3
+sparse_multi_token_attention,liger,full,memory,MB,L,sequence length,256,20.005859375,20.005859375,20.005859375,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:52,0.6.3
+sparse_multi_token_attention,liger,full,memory,MB,L,sequence length,512,80.005859375,80.005859375,80.005859375,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:52,0.6.3
+sparse_multi_token_attention,torch,full,memory,MB,L,sequence length,32,0.447265625,0.447265625,0.447265625,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:52,0.6.3
+sparse_multi_token_attention,torch,full,memory,MB,L,sequence length,64,1.7705078125,1.7705078125,1.7705078125,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:52,0.6.3
+sparse_multi_token_attention,torch,full,memory,MB,L,sequence length,128,7.04833984375,7.04833984375,7.04833984375,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:52,0.6.3
+sparse_multi_token_attention,torch,full,memory,MB,L,sequence length,256,28.15478515625,28.15478515625,28.15478515625,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:52,0.6.3
+sparse_multi_token_attention,torch,full,memory,MB,L,sequence length,512,112.55517578125,112.55517578125,112.55517578125,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 00:59:52,0.6.3
+sparsemax,liger,forward,speed,ms,V,feature size,1024,0.09571199864149094,0.09489919990301132,0.09644799679517746,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:00,0.6.3
+sparsemax,liger,forward,speed,ms,V,feature size,2048,0.1844159960746765,0.18355199694633484,0.18534399569034576,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:00,0.6.3
+sparsemax,liger,forward,speed,ms,V,feature size,4096,0.3524479866027832,0.35139200091362,0.3536768019199371,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:00,0.6.3
+sparsemax,liger,forward,speed,ms,V,feature size,8192,1.5446079969406128,1.5434367895126342,1.5466560125350952,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:00,0.6.3
+sparsemax,liger,forward,speed,ms,V,feature size,16384,3.002959966659546,3.0006848812103275,3.004512071609497,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:00,0.6.3
+sparsemax,liger,forward,speed,ms,V,feature size,32768,5.964143991470337,5.961670589447021,5.966809558868408,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:00,0.6.3
+sparsemax,torch,forward,speed,ms,V,feature size,1024,0.21347199380397797,0.2126079946756363,0.21430400013923645,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:04,0.6.3
+sparsemax,torch,forward,speed,ms,V,feature size,2048,0.3912320137023926,0.390175998210907,0.3922944009304047,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:04,0.6.3
+sparsemax,torch,forward,speed,ms,V,feature size,4096,0.6871199905872345,0.6855999827384949,0.6885759830474854,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:04,0.6.3
+sparsemax,torch,forward,speed,ms,V,feature size,8192,2.1348800659179688,2.133023977279663,2.1372672080993653,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:04,0.6.3
+sparsemax,torch,forward,speed,ms,V,feature size,16384,4.015520095825195,4.012991905212402,4.017695903778076,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:04,0.6.3
+sparsemax,torch,forward,speed,ms,V,feature size,32768,7.866527795791626,7.864134502410889,7.8687872886657715,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:04,0.6.3
+sparsemax,liger,full,speed,ms,V,feature size,1024,0.23969599604606628,0.20442240238189696,0.2999616086483002,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:08,0.6.3
+sparsemax,liger,full,speed,ms,V,feature size,2048,0.2287360057234764,0.20237440168857573,0.27928320765495296,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:08,0.6.3
+sparsemax,liger,full,speed,ms,V,feature size,4096,0.3927040100097656,0.3917759954929352,0.3936319947242737,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:08,0.6.3
+sparsemax,liger,full,speed,ms,V,feature size,8192,1.632383942604065,1.629254412651062,1.6350336313247682,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:08,0.6.3
+sparsemax,liger,full,speed,ms,V,feature size,16384,3.1637439727783203,3.161574363708496,3.165753650665283,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:08,0.6.3
+sparsemax,liger,full,speed,ms,V,feature size,32768,6.353024005889893,6.3500288963317875,6.355475139617919,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:08,0.6.3
+sparsemax,torch,full,speed,ms,V,feature size,1024,0.3811040073633194,0.35471999645233154,0.4280639886856079,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:11,0.6.3
+sparsemax,torch,full,speed,ms,V,feature size,2048,0.5490559935569763,0.547648012638092,0.5506880283355713,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:11,0.6.3
+sparsemax,torch,full,speed,ms,V,feature size,4096,1.0326080322265625,1.0309760570526123,1.0344640016555786,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:11,0.6.3
+sparsemax,torch,full,speed,ms,V,feature size,8192,2.8065919876098633,2.803430366516113,2.809446430206299,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:11,0.6.3
+sparsemax,torch,full,speed,ms,V,feature size,16384,5.319311857223511,5.316537666320801,5.322207927703857,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:11,0.6.3
+sparsemax,torch,full,speed,ms,V,feature size,32768,10.42467212677002,10.420127868652344,10.4269437789917,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:11,0.6.3
+sparsemax,liger,backward,speed,ms,V,feature size,1024,0.07548800110816956,0.06302080303430557,0.09097599983215332,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:14,0.6.3
+sparsemax,liger,backward,speed,ms,V,feature size,2048,0.07344000041484833,0.06217600032687187,0.08851200342178345,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:14,0.6.3
+sparsemax,liger,backward,speed,ms,V,feature size,4096,0.07254400104284286,0.06190719977021217,0.08676479756832123,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:14,0.6.3
+sparsemax,liger,backward,speed,ms,V,feature size,8192,0.09545599669218063,0.0910400003194809,0.09724800288677216,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:14,0.6.3
+sparsemax,liger,backward,speed,ms,V,feature size,16384,0.16484799981117249,0.16380800306797028,0.16582399606704712,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:14,0.6.3
+sparsemax,liger,backward,speed,ms,V,feature size,32768,0.3949440121650696,0.3940480053424835,0.39587199687957764,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:14,0.6.3
+sparsemax,torch,backward,speed,ms,V,feature size,1024,0.13180799782276154,0.11214719712734222,0.15341440439224244,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:17,0.6.3
+sparsemax,torch,backward,speed,ms,V,feature size,2048,0.16553600132465363,0.16451199352741241,0.16678400337696075,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:17,0.6.3
+sparsemax,torch,backward,speed,ms,V,feature size,4096,0.34668800234794617,0.34552319049835206,0.3479488074779511,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:17,0.6.3
+sparsemax,torch,backward,speed,ms,V,feature size,8192,0.6735680103302002,0.6719679832458496,0.6749631881713867,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:17,0.6.3
+sparsemax,torch,backward,speed,ms,V,feature size,16384,1.3071680068969727,1.3057279586791992,1.3086400032043457,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:17,0.6.3
+sparsemax,torch,backward,speed,ms,V,feature size,32768,2.557471990585327,2.554719924926758,2.559583902359009,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:17,0.6.3
+sparsemax,liger,full,memory,MB,V,feature size,1024,48.0078125,48.0078125,48.0078125,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:17,0.6.3
+sparsemax,liger,full,memory,MB,V,feature size,2048,96.015625,96.015625,96.015625,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:17,0.6.3
+sparsemax,liger,full,memory,MB,V,feature size,4096,192.03125,192.03125,192.03125,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:17,0.6.3
+sparsemax,liger,full,memory,MB,V,feature size,8192,704.00048828125,704.00048828125,704.00048828125,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:17,0.6.3
+sparsemax,liger,full,memory,MB,V,feature size,16384,1408.00048828125,1408.00048828125,1408.00048828125,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:17,0.6.3
+sparsemax,liger,full,memory,MB,V,feature size,32768,2816.00048828125,2816.00048828125,2816.00048828125,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:17,0.6.3
+sparsemax,torch,full,memory,MB,V,feature size,1024,82.03515625,82.03515625,82.03515625,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:17,0.6.3
+sparsemax,torch,full,memory,MB,V,feature size,2048,164.0390625,164.0390625,164.0390625,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:17,0.6.3
+sparsemax,torch,full,memory,MB,V,feature size,4096,328.046875,328.046875,328.046875,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:17,0.6.3
+sparsemax,torch,full,memory,MB,V,feature size,8192,704.00048828125,704.00048828125,704.00048828125,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:17,0.6.3
+sparsemax,torch,full,memory,MB,V,feature size,16384,1408.00048828125,1408.00048828125,1408.00048828125,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:17,0.6.3
+sparsemax,torch,full,memory,MB,V,feature size,32768,2816.00048828125,2816.00048828125,2816.00048828125,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:17,0.6.3
+swiglu,liger,forward,speed,ms,T,sequence length,1024,1.484287977218628,1.4799295663833618,1.4849792003631592,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:28,0.6.3
+swiglu,liger,forward,speed,ms,T,sequence length,2048,3.0234880447387695,3.0222400665283202,3.0281151771545414,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:28,0.6.3
+swiglu,liger,forward,speed,ms,T,sequence length,4096,6.116864204406738,6.116864204406738,6.116864204406738,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:28,0.6.3
+swiglu,liger,forward,speed,ms,T,sequence length,8192,13.27667236328125,13.27667236328125,13.27667236328125,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:28,0.6.3
+swiglu,huggingface,forward,speed,ms,T,sequence length,1024,1.5953439474105835,1.5938559770584106,1.5957119464874268,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:31,0.6.3
+swiglu,huggingface,forward,speed,ms,T,sequence length,2048,3.0002241134643555,2.999840021133423,3.0019905090332033,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:31,0.6.3
+swiglu,huggingface,forward,speed,ms,T,sequence length,4096,6.631552219390869,6.631552219390869,6.631552219390869,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:31,0.6.3
+swiglu,huggingface,forward,speed,ms,T,sequence length,8192,13.44934368133545,13.44934368133545,13.44934368133545,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:31,0.6.3
+swiglu,liger,full,speed,ms,T,sequence length,1024,5.2791359424591064,5.279001617431641,5.279270267486572,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:35,0.6.3
+swiglu,liger,full,speed,ms,T,sequence length,2048,10.193856239318848,10.193856239318848,10.193856239318848,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:35,0.6.3
+swiglu,liger,full,speed,ms,T,sequence length,4096,20.07948875427246,20.07948875427246,20.07948875427246,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:35,0.6.3
+swiglu,liger,full,speed,ms,T,sequence length,8192,39.3155517578125,39.3155517578125,39.3155517578125,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:35,0.6.3
+swiglu,huggingface,full,speed,ms,T,sequence length,1024,4.971807956695557,4.971807956695557,4.971807956695557,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:38,0.6.3
+swiglu,huggingface,full,speed,ms,T,sequence length,2048,9.949983596801758,9.949983596801758,9.949983596801758,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:38,0.6.3
+swiglu,huggingface,full,speed,ms,T,sequence length,4096,21.228927612304688,21.228927612304688,21.228927612304688,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:38,0.6.3
+swiglu,huggingface,full,speed,ms,T,sequence length,8192,40.17123031616211,40.17123031616211,40.17123031616211,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:38,0.6.3
+swiglu,liger,backward,speed,ms,T,sequence length,1024,3.4129440784454346,3.4128576755523685,3.413030481338501,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:41,0.6.3
+swiglu,liger,backward,speed,ms,T,sequence length,2048,6.154111862182617,6.154111862182617,6.154111862182617,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:41,0.6.3
+swiglu,liger,backward,speed,ms,T,sequence length,4096,13.242655754089355,13.242655754089355,13.242655754089355,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:41,0.6.3
+swiglu,liger,backward,speed,ms,T,sequence length,8192,26.559776306152344,26.559776306152344,26.559776306152344,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:41,0.6.3
+swiglu,huggingface,backward,speed,ms,T,sequence length,1024,3.5074559450149536,3.506246376037598,3.5086655139923097,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:45,0.6.3
+swiglu,huggingface,backward,speed,ms,T,sequence length,2048,7.002272129058838,7.002272129058838,7.002272129058838,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:45,0.6.3
+swiglu,huggingface,backward,speed,ms,T,sequence length,4096,12.943391799926758,12.943391799926758,12.943391799926758,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:45,0.6.3
+swiglu,huggingface,backward,speed,ms,T,sequence length,8192,26.776607513427734,26.776607513427734,26.776607513427734,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:45,0.6.3
+swiglu,liger,full,memory,MB,T,sequence length,1024,1148.0,1148.0,1148.0,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:47,0.6.3
+swiglu,liger,full,memory,MB,T,sequence length,2048,1630.0,1630.0,1630.0,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:47,0.6.3
+swiglu,liger,full,memory,MB,T,sequence length,4096,2594.0,2594.0,2594.0,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:47,0.6.3
+swiglu,liger,full,memory,MB,T,sequence length,8192,4522.0,4522.0,4522.0,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:47,0.6.3
+swiglu,huggingface,full,memory,MB,T,sequence length,1024,1342.0,1342.0,1342.0,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:50,0.6.3
+swiglu,huggingface,full,memory,MB,T,sequence length,2048,2040.0,2040.0,2040.0,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:50,0.6.3
+swiglu,huggingface,full,memory,MB,T,sequence length,4096,3500.0,3500.0,3500.0,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:50,0.6.3
+swiglu,huggingface,full,memory,MB,T,sequence length,8192,6420.0,6420.0,6420.0,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:50,0.6.3
+tiled_geglu,huggingface,full,speed,ms,T,sequence length,1024,0.5910720229148865,0.5906495928764344,0.5924160242080689,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:58,0.6.3
+tiled_geglu,huggingface,full,speed,ms,T,sequence length,2048,1.05840003490448,1.0580799579620361,1.0595840215682983,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:58,0.6.3
+tiled_geglu,huggingface,full,speed,ms,T,sequence length,4096,1.9843520522117615,1.9830912351608276,1.9895168304443358,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:58,0.6.3
+tiled_geglu,huggingface,full,speed,ms,T,sequence length,8192,3.8954720497131348,3.8915264129638674,3.899417686462402,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:58,0.6.3
+tiled_geglu,huggingface,full,speed,ms,T,sequence length,16384,7.743807792663574,7.743807792663574,7.743807792663574,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:00:58,0.6.3
+tiled_geglu,liger,full,speed,ms,T,sequence length,1024,0.5666239857673645,0.5489344120025634,0.5847936034202577,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:00,0.6.3
+tiled_geglu,liger,full,speed,ms,T,sequence length,2048,0.9850239753723145,0.9841088175773621,0.986624002456665,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:00,0.6.3
+tiled_geglu,liger,full,speed,ms,T,sequence length,4096,1.8743679523468018,1.8738368034362793,1.8772032499313354,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:00,0.6.3
+tiled_geglu,liger,full,speed,ms,T,sequence length,8192,3.625391960144043,3.623443174362183,3.6273407459259035,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:00,0.6.3
+tiled_geglu,liger,full,speed,ms,T,sequence length,16384,7.55244779586792,7.55244779586792,7.55244779586792,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:00,0.6.3
+tiled_geglu,liger_tiled,full,speed,ms,T,sequence length,1024,2.526304006576538,2.5108672142028805,2.531103944778442,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:01,0.6.3
+tiled_geglu,liger_tiled,full,speed,ms,T,sequence length,2048,2.746272087097168,2.7324288845062252,2.7466943740844725,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:01,0.6.3
+tiled_geglu,liger_tiled,full,speed,ms,T,sequence length,4096,3.0465281009674072,3.042937660217285,3.0473536491394047,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:01,0.6.3
+tiled_geglu,liger_tiled,full,speed,ms,T,sequence length,8192,5.720928192138672,5.720928192138672,5.720928192138672,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:01,0.6.3
+tiled_geglu,liger_tiled,full,speed,ms,T,sequence length,16384,11.121408462524414,11.121408462524414,11.121408462524414,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:01,0.6.3
+tiled_geglu,deepspeed_tiled,full,speed,ms,T,sequence length,1024,2.034832000732422,2.029420804977417,2.0369920253753664,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:02,0.6.3
+tiled_geglu,deepspeed_tiled,full,speed,ms,T,sequence length,2048,2.235919952392578,2.179833507537842,2.283852767944336,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:02,0.6.3
+tiled_geglu,deepspeed_tiled,full,speed,ms,T,sequence length,4096,3.2293280363082886,3.218182420730591,3.2404736518859867,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:02,0.6.3
+tiled_geglu,deepspeed_tiled,full,speed,ms,T,sequence length,8192,5.880671977996826,5.880671977996826,5.880671977996826,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:02,0.6.3
+tiled_geglu,deepspeed_tiled,full,speed,ms,T,sequence length,16384,11.913311958312988,11.913311958312988,11.913311958312988,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:02,0.6.3
+tiled_geglu,huggingface,forward,speed,ms,T,sequence length,1024,0.1735839992761612,0.17325439751148225,0.17420800030231476,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:03,0.6.3
+tiled_geglu,huggingface,forward,speed,ms,T,sequence length,2048,0.3309440016746521,0.32916479706764223,0.3317183971405029,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:03,0.6.3
+tiled_geglu,huggingface,forward,speed,ms,T,sequence length,4096,0.6414399743080139,0.6407487869262696,0.6423935890197754,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:03,0.6.3
+tiled_geglu,huggingface,forward,speed,ms,T,sequence length,8192,1.2742719650268555,1.2734591960906982,1.2760511636734009,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:03,0.6.3
+tiled_geglu,huggingface,forward,speed,ms,T,sequence length,16384,2.5227200984954834,2.521894407272339,2.5235456466674804,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:03,0.6.3
+tiled_geglu,liger,forward,speed,ms,T,sequence length,1024,0.16278400272130966,0.16150399446487426,0.1630784034729004,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:04,0.6.3
+tiled_geglu,liger,forward,speed,ms,T,sequence length,2048,0.30748799443244934,0.30586240291595457,0.3082687914371491,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:04,0.6.3
+tiled_geglu,liger,forward,speed,ms,T,sequence length,4096,0.6125119924545288,0.6106751918792725,0.6138816118240357,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:04,0.6.3
+tiled_geglu,liger,forward,speed,ms,T,sequence length,8192,1.1862399578094482,1.1860608339309693,1.1878080129623414,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:04,0.6.3
+tiled_geglu,liger,forward,speed,ms,T,sequence length,16384,2.344512104988098,2.342124891281128,2.347424030303955,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:04,0.6.3
+tiled_geglu,liger_tiled,forward,speed,ms,T,sequence length,1024,0.5095040202140808,0.49159039855003356,0.5393664121627808,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:05,0.6.3
+tiled_geglu,liger_tiled,forward,speed,ms,T,sequence length,2048,0.49099200963974,0.4774079918861389,0.5068479776382446,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:05,0.6.3
+tiled_geglu,liger_tiled,forward,speed,ms,T,sequence length,4096,0.7765439748764038,0.7758079767227173,0.7770559787750244,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:05,0.6.3
+tiled_geglu,liger_tiled,forward,speed,ms,T,sequence length,8192,1.466111958026886,1.465343952178955,1.4666240215301514,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:05,0.6.3
+tiled_geglu,liger_tiled,forward,speed,ms,T,sequence length,16384,2.875744104385376,2.8730751991271974,2.8982656478881834,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:05,0.6.3
+tiled_geglu,deepspeed_tiled,forward,speed,ms,T,sequence length,1024,0.3561599999666214,0.33852160573005674,0.37421441078186035,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:05,0.6.3
+tiled_geglu,deepspeed_tiled,forward,speed,ms,T,sequence length,2048,0.4638880044221878,0.4630400002002716,0.4644991993904114,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:05,0.6.3
+tiled_geglu,deepspeed_tiled,forward,speed,ms,T,sequence length,4096,0.8218240141868591,0.8204159736633301,0.8230080008506775,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:05,0.6.3
+tiled_geglu,deepspeed_tiled,forward,speed,ms,T,sequence length,8192,1.5649279952049255,1.564255952835083,1.5660480260849,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:05,0.6.3
+tiled_geglu,deepspeed_tiled,forward,speed,ms,T,sequence length,16384,3.090303897857666,3.0607744216918946,3.11407356262207,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:05,0.6.3
+tiled_geglu,huggingface,backward,speed,ms,T,sequence length,1024,0.4069119989871979,0.40657278895378113,0.4076608002185822,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:06,0.6.3
+tiled_geglu,huggingface,backward,speed,ms,T,sequence length,2048,0.7094720005989075,0.7077888131141662,0.7102272152900696,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:06,0.6.3
+tiled_geglu,huggingface,backward,speed,ms,T,sequence length,4096,1.3466560244560242,1.3188480138778687,1.3491519689559937,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:06,0.6.3
+tiled_geglu,huggingface,backward,speed,ms,T,sequence length,8192,2.541951894760132,2.53953275680542,2.544486331939697,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:06,0.6.3
+tiled_geglu,huggingface,backward,speed,ms,T,sequence length,16384,5.235968112945557,5.235968112945557,5.235968112945557,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:06,0.6.3
+tiled_geglu,liger,backward,speed,ms,T,sequence length,1024,0.36905598640441895,0.36819198727607727,0.3701440095901489,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:07,0.6.3
+tiled_geglu,liger,backward,speed,ms,T,sequence length,2048,0.6517760157585144,0.6493312001228333,0.6535167932510376,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:07,0.6.3
+tiled_geglu,liger,backward,speed,ms,T,sequence length,4096,1.2178879976272583,1.2165504455566407,1.2187584161758422,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:07,0.6.3
+tiled_geglu,liger,backward,speed,ms,T,sequence length,8192,2.434272050857544,2.4332543849945067,2.435807991027832,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:07,0.6.3
+tiled_geglu,liger,backward,speed,ms,T,sequence length,16384,4.8334879875183105,4.833247947692871,4.83372802734375,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:07,0.6.3
+tiled_geglu,liger_tiled,backward,speed,ms,T,sequence length,1024,1.7984639406204224,1.7836095571517945,1.831231951713562,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:08,0.6.3
+tiled_geglu,liger_tiled,backward,speed,ms,T,sequence length,2048,1.748960018157959,1.7273920059204102,1.8004351615905763,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:08,0.6.3
+tiled_geglu,liger_tiled,backward,speed,ms,T,sequence length,4096,2.202415943145752,2.20196475982666,2.202828788757324,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:08,0.6.3
+tiled_geglu,liger_tiled,backward,speed,ms,T,sequence length,8192,3.922096014022827,3.9171711921691896,3.9270208358764647,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:08,0.6.3
+tiled_geglu,liger_tiled,backward,speed,ms,T,sequence length,16384,7.443488121032715,7.443488121032715,7.443488121032715,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:08,0.6.3
+tiled_geglu,deepspeed_tiled,backward,speed,ms,T,sequence length,1024,1.4688160419464111,1.447167992591858,1.5628160238265991,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:09,0.6.3
+tiled_geglu,deepspeed_tiled,backward,speed,ms,T,sequence length,2048,1.408191978931427,1.406816005706787,1.4097599983215332,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:09,0.6.3
+tiled_geglu,deepspeed_tiled,backward,speed,ms,T,sequence length,4096,2.357151985168457,2.356761646270752,2.3581567764282223,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:09,0.6.3
+tiled_geglu,deepspeed_tiled,backward,speed,ms,T,sequence length,8192,4.2194719314575195,4.218195056915284,4.220748805999756,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:09,0.6.3
+tiled_geglu,deepspeed_tiled,backward,speed,ms,T,sequence length,16384,8.308480262756348,8.308480262756348,8.308480262756348,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:09,0.6.3
+tiled_geglu,huggingface,full,memory,MB,T,sequence length,1024,312.0,312.0,312.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:10,0.6.3
+tiled_geglu,huggingface,full,memory,MB,T,sequence length,2048,448.0,448.0,448.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:10,0.6.3
+tiled_geglu,huggingface,full,memory,MB,T,sequence length,4096,736.0,736.0,736.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:10,0.6.3
+tiled_geglu,huggingface,full,memory,MB,T,sequence length,8192,1312.0,1312.0,1312.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:10,0.6.3
+tiled_geglu,huggingface,full,memory,MB,T,sequence length,16384,2464.0,2464.0,2464.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:10,0.6.3
+tiled_geglu,liger,full,memory,MB,T,sequence length,1024,280.0,280.0,280.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:10,0.6.3
+tiled_geglu,liger,full,memory,MB,T,sequence length,2048,384.0,384.0,384.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:10,0.6.3
+tiled_geglu,liger,full,memory,MB,T,sequence length,4096,592.0,592.0,592.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:10,0.6.3
+tiled_geglu,liger,full,memory,MB,T,sequence length,8192,1008.0,1008.0,1008.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:10,0.6.3
+tiled_geglu,liger,full,memory,MB,T,sequence length,16384,1840.0,1840.0,1840.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:10,0.6.3
+tiled_geglu,liger_tiled,full,memory,MB,T,sequence length,1024,234.0,234.0,234.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:11,0.6.3
+tiled_geglu,liger_tiled,full,memory,MB,T,sequence length,2048,292.0,292.0,292.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:11,0.6.3
+tiled_geglu,liger_tiled,full,memory,MB,T,sequence length,4096,408.0,408.0,408.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:11,0.6.3
+tiled_geglu,liger_tiled,full,memory,MB,T,sequence length,8192,640.0,640.0,640.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:11,0.6.3
+tiled_geglu,liger_tiled,full,memory,MB,T,sequence length,16384,1104.0,1104.0,1104.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:11,0.6.3
+tiled_geglu,deepspeed_tiled,full,memory,MB,T,sequence length,1024,238.0,238.0,238.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:12,0.6.3
+tiled_geglu,deepspeed_tiled,full,memory,MB,T,sequence length,2048,300.0,300.0,300.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:12,0.6.3
+tiled_geglu,deepspeed_tiled,full,memory,MB,T,sequence length,4096,424.0,424.0,424.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:12,0.6.3
+tiled_geglu,deepspeed_tiled,full,memory,MB,T,sequence length,8192,688.0,688.0,688.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:12,0.6.3
+tiled_geglu,deepspeed_tiled,full,memory,MB,T,sequence length,16384,1216.0,1216.0,1216.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:12,0.6.3
+tiled_geglu,huggingface,forward,memory,MB,T,sequence length,1024,192.0,192.0,192.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:13,0.6.3
+tiled_geglu,huggingface,forward,memory,MB,T,sequence length,2048,272.0,272.0,272.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:13,0.6.3
+tiled_geglu,huggingface,forward,memory,MB,T,sequence length,4096,432.0,432.0,432.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:13,0.6.3
+tiled_geglu,huggingface,forward,memory,MB,T,sequence length,8192,752.0,752.0,752.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:13,0.6.3
+tiled_geglu,huggingface,forward,memory,MB,T,sequence length,16384,1392.0,1392.0,1392.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:13,0.6.3
+tiled_geglu,liger,forward,memory,MB,T,sequence length,1024,176.0,176.0,176.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:13,0.6.3
+tiled_geglu,liger,forward,memory,MB,T,sequence length,2048,240.0,240.0,240.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:13,0.6.3
+tiled_geglu,liger,forward,memory,MB,T,sequence length,4096,368.0,368.0,368.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:13,0.6.3
+tiled_geglu,liger,forward,memory,MB,T,sequence length,8192,624.0,624.0,624.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:13,0.6.3
+tiled_geglu,liger,forward,memory,MB,T,sequence length,16384,1136.0,1136.0,1136.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:13,0.6.3
+tiled_geglu,liger_tiled,forward,memory,MB,T,sequence length,1024,140.0,140.0,140.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:14,0.6.3
+tiled_geglu,liger_tiled,forward,memory,MB,T,sequence length,2048,168.0,168.0,168.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:14,0.6.3
+tiled_geglu,liger_tiled,forward,memory,MB,T,sequence length,4096,224.0,224.0,224.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:14,0.6.3
+tiled_geglu,liger_tiled,forward,memory,MB,T,sequence length,8192,336.0,336.0,336.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:14,0.6.3
+tiled_geglu,liger_tiled,forward,memory,MB,T,sequence length,16384,560.0,560.0,560.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:14,0.6.3
+tiled_geglu,deepspeed_tiled,forward,memory,MB,T,sequence length,1024,138.0,138.0,138.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:15,0.6.3
+tiled_geglu,deepspeed_tiled,forward,memory,MB,T,sequence length,2048,164.0,164.0,164.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:15,0.6.3
+tiled_geglu,deepspeed_tiled,forward,memory,MB,T,sequence length,4096,216.0,216.0,216.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:15,0.6.3
+tiled_geglu,deepspeed_tiled,forward,memory,MB,T,sequence length,8192,320.0,320.0,320.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:15,0.6.3
+tiled_geglu,deepspeed_tiled,forward,memory,MB,T,sequence length,16384,528.0,528.0,528.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:15,0.6.3
+tiled_geglu,huggingface,backward,memory,MB,T,sequence length,1024,312.0,312.0,312.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:15,0.6.3
+tiled_geglu,huggingface,backward,memory,MB,T,sequence length,2048,448.0,448.0,448.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:15,0.6.3
+tiled_geglu,huggingface,backward,memory,MB,T,sequence length,4096,736.0,736.0,736.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:15,0.6.3
+tiled_geglu,huggingface,backward,memory,MB,T,sequence length,8192,1312.0,1312.0,1312.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:15,0.6.3
+tiled_geglu,huggingface,backward,memory,MB,T,sequence length,16384,2464.0,2464.0,2464.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:15,0.6.3
+tiled_geglu,liger,backward,memory,MB,T,sequence length,1024,280.0,280.0,280.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:16,0.6.3
+tiled_geglu,liger,backward,memory,MB,T,sequence length,2048,384.0,384.0,384.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:16,0.6.3
+tiled_geglu,liger,backward,memory,MB,T,sequence length,4096,592.0,592.0,592.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:16,0.6.3
+tiled_geglu,liger,backward,memory,MB,T,sequence length,8192,1008.0,1008.0,1008.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:16,0.6.3
+tiled_geglu,liger,backward,memory,MB,T,sequence length,16384,1840.0,1840.0,1840.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:16,0.6.3
+tiled_geglu,liger_tiled,backward,memory,MB,T,sequence length,1024,234.0,234.0,234.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:17,0.6.3
+tiled_geglu,liger_tiled,backward,memory,MB,T,sequence length,2048,292.0,292.0,292.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:17,0.6.3
+tiled_geglu,liger_tiled,backward,memory,MB,T,sequence length,4096,408.0,408.0,408.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:17,0.6.3
+tiled_geglu,liger_tiled,backward,memory,MB,T,sequence length,8192,640.0,640.0,640.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:17,0.6.3
+tiled_geglu,liger_tiled,backward,memory,MB,T,sequence length,16384,1104.0,1104.0,1104.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:17,0.6.3
+tiled_geglu,deepspeed_tiled,backward,memory,MB,T,sequence length,1024,238.0,238.0,238.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:17,0.6.3
+tiled_geglu,deepspeed_tiled,backward,memory,MB,T,sequence length,2048,300.0,300.0,300.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:17,0.6.3
+tiled_geglu,deepspeed_tiled,backward,memory,MB,T,sequence length,4096,424.0,424.0,424.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:17,0.6.3
+tiled_geglu,deepspeed_tiled,backward,memory,MB,T,sequence length,8192,688.0,688.0,688.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:17,0.6.3
+tiled_geglu,deepspeed_tiled,backward,memory,MB,T,sequence length,16384,1216.0,1216.0,1216.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""gelu_pytorch_tanh"", ""activation_type"": ""geglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:17,0.6.3
+tiled_swiglu,huggingface,full,speed,ms,T,sequence length,1024,0.5901600122451782,0.5889151930809021,0.5912064075469972,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:18,0.6.3
+tiled_swiglu,huggingface,full,speed,ms,T,sequence length,2048,1.060368001461029,1.059500813484192,1.0610176086425782,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:18,0.6.3
+tiled_swiglu,huggingface,full,speed,ms,T,sequence length,4096,2.0172159671783447,2.0164095401763915,2.018444776535034,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:18,0.6.3
+tiled_swiglu,huggingface,full,speed,ms,T,sequence length,8192,3.907663941383362,3.9068095684051514,3.9085183143615723,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:18,0.6.3
+tiled_swiglu,huggingface,full,speed,ms,T,sequence length,16384,7.903168201446533,7.903168201446533,7.903168201446533,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:18,0.6.3
+tiled_swiglu,liger,full,speed,ms,T,sequence length,1024,0.5512320101261139,0.5491391777992248,0.5672832131385803,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:19,0.6.3
+tiled_swiglu,liger,full,speed,ms,T,sequence length,2048,0.9856640100479126,0.9840384006500243,0.9982655644416809,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:19,0.6.3
+tiled_swiglu,liger,full,speed,ms,T,sequence length,4096,1.854848027229309,1.851910400390625,1.8552895545959471,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:19,0.6.3
+tiled_swiglu,liger,full,speed,ms,T,sequence length,8192,3.767840027809143,3.7671680450439458,3.768512010574341,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:19,0.6.3
+tiled_swiglu,liger,full,speed,ms,T,sequence length,16384,7.89734411239624,7.89734411239624,7.89734411239624,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:19,0.6.3
+tiled_swiglu,liger_tiled,full,speed,ms,T,sequence length,1024,2.4112000465393066,2.2762431621551515,2.4264640331268312,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:20,0.6.3
+tiled_swiglu,liger_tiled,full,speed,ms,T,sequence length,2048,2.3971519470214844,2.3185855388641357,2.437606382369995,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:20,0.6.3
+tiled_swiglu,liger_tiled,full,speed,ms,T,sequence length,4096,3.0018880367279053,2.9998911857604984,3.0021951675415037,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:20,0.6.3
+tiled_swiglu,liger_tiled,full,speed,ms,T,sequence length,8192,5.623199939727783,5.623199939727783,5.623199939727783,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:20,0.6.3
+tiled_swiglu,liger_tiled,full,speed,ms,T,sequence length,16384,11.214240074157715,11.214240074157715,11.214240074157715,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:20,0.6.3
+tiled_swiglu,deepspeed_tiled,full,speed,ms,T,sequence length,1024,1.8433120250701904,1.7907776117324827,1.8835456132888795,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:21,0.6.3
+tiled_swiglu,deepspeed_tiled,full,speed,ms,T,sequence length,2048,2.0497759580612183,2.0242176055908203,2.076051187515259,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:21,0.6.3
+tiled_swiglu,deepspeed_tiled,full,speed,ms,T,sequence length,4096,3.209215998649597,3.2081024169921877,3.210329580307007,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:21,0.6.3
+tiled_swiglu,deepspeed_tiled,full,speed,ms,T,sequence length,8192,6.134399890899658,6.134399890899658,6.134399890899658,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:21,0.6.3
+tiled_swiglu,deepspeed_tiled,full,speed,ms,T,sequence length,16384,11.789376258850098,11.789376258850098,11.789376258850098,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:21,0.6.3
+tiled_swiglu,huggingface,forward,speed,ms,T,sequence length,1024,0.17265599966049194,0.17231999337673187,0.17310720384120942,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:22,0.6.3
+tiled_swiglu,huggingface,forward,speed,ms,T,sequence length,2048,0.3301759958267212,0.3289471983909607,0.33189759850502015,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:22,0.6.3
+tiled_swiglu,huggingface,forward,speed,ms,T,sequence length,4096,0.6412799954414368,0.6402687907218934,0.6417216062545776,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:22,0.6.3
+tiled_swiglu,huggingface,forward,speed,ms,T,sequence length,8192,1.2761919498443604,1.2757503986358643,1.276409602165222,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:22,0.6.3
+tiled_swiglu,huggingface,forward,speed,ms,T,sequence length,16384,2.523519992828369,2.5224639892578127,2.524019241333008,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:22,0.6.3
+tiled_swiglu,liger,forward,speed,ms,T,sequence length,1024,0.1626719981431961,0.16100480258464814,0.16314240396022794,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:23,0.6.3
+tiled_swiglu,liger,forward,speed,ms,T,sequence length,2048,0.3078080117702484,0.30674560070037843,0.3092480003833771,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:23,0.6.3
+tiled_swiglu,liger,forward,speed,ms,T,sequence length,4096,0.5929760038852692,0.5920064091682434,0.594809603691101,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:23,0.6.3
+tiled_swiglu,liger,forward,speed,ms,T,sequence length,8192,1.187615990638733,1.1871999979019165,1.1894015789031982,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:23,0.6.3
+tiled_swiglu,liger,forward,speed,ms,T,sequence length,16384,2.3486560583114624,2.3469377040863035,2.350451183319092,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:23,0.6.3
+tiled_swiglu,liger_tiled,forward,speed,ms,T,sequence length,1024,0.4979200065135956,0.47706241011619566,0.5161088109016418,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:24,0.6.3
+tiled_swiglu,liger_tiled,forward,speed,ms,T,sequence length,2048,0.501440018415451,0.48080000281333923,0.5235840082168579,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:24,0.6.3
+tiled_swiglu,liger_tiled,forward,speed,ms,T,sequence length,4096,0.7755200266838074,0.7747200131416321,0.777184009552002,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:24,0.6.3
+tiled_swiglu,liger_tiled,forward,speed,ms,T,sequence length,8192,1.5104320049285889,1.4875839948654175,1.5106240510940552,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:24,0.6.3
+tiled_swiglu,liger_tiled,forward,speed,ms,T,sequence length,16384,2.8447999954223633,2.8445120334625242,2.84781436920166,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:24,0.6.3
+tiled_swiglu,deepspeed_tiled,forward,speed,ms,T,sequence length,1024,0.36875200271606445,0.3557247936725616,0.4026367962360382,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:25,0.6.3
+tiled_swiglu,deepspeed_tiled,forward,speed,ms,T,sequence length,2048,0.4625280052423477,0.4598400056362152,0.46446719765663147,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:25,0.6.3
+tiled_swiglu,deepspeed_tiled,forward,speed,ms,T,sequence length,4096,0.8161919713020325,0.815168023109436,0.8171839714050293,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:25,0.6.3
+tiled_swiglu,deepspeed_tiled,forward,speed,ms,T,sequence length,8192,1.5665919780731201,1.5652159452438354,1.5688960552215576,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:25,0.6.3
+tiled_swiglu,deepspeed_tiled,forward,speed,ms,T,sequence length,16384,3.0216639041900635,3.0201279640197756,3.0225279331207275,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:25,0.6.3
+tiled_swiglu,huggingface,backward,speed,ms,T,sequence length,1024,0.40726399421691895,0.40652158856391907,0.4079103946685791,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:25,0.6.3
+tiled_swiglu,huggingface,backward,speed,ms,T,sequence length,2048,0.711679995059967,0.7112704157829285,0.7130240082740784,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:25,0.6.3
+tiled_swiglu,huggingface,backward,speed,ms,T,sequence length,4096,1.3157600164413452,1.3140159845352173,1.3172160387039185,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:25,0.6.3
+tiled_swiglu,huggingface,backward,speed,ms,T,sequence length,8192,2.5915839672088623,2.589049530029297,2.59214072227478,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:25,0.6.3
+tiled_swiglu,huggingface,backward,speed,ms,T,sequence length,16384,4.997663974761963,4.997663974761963,4.997663974761963,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:25,0.6.3
+tiled_swiglu,liger,backward,speed,ms,T,sequence length,1024,0.3721279948949814,0.37155839800834656,0.37303040623664857,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:26,0.6.3
+tiled_swiglu,liger,backward,speed,ms,T,sequence length,2048,0.650111973285675,0.6467456102371216,0.6523711800575256,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:26,0.6.3
+tiled_swiglu,liger,backward,speed,ms,T,sequence length,4096,1.2208960056304932,1.2197696447372437,1.2235584020614625,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:26,0.6.3
+tiled_swiglu,liger,backward,speed,ms,T,sequence length,8192,2.3988161087036133,2.3981440544128416,2.400524854660034,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:26,0.6.3
+tiled_swiglu,liger,backward,speed,ms,T,sequence length,16384,4.686511993408203,4.685638523101806,4.6873854637146,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:26,0.6.3
+tiled_swiglu,liger_tiled,backward,speed,ms,T,sequence length,1024,1.6261440515518188,1.5320127725601196,1.6904831886291503,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:27,0.6.3
+tiled_swiglu,liger_tiled,backward,speed,ms,T,sequence length,2048,1.6602239608764648,1.6454784393310549,1.6883456468582154,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:27,0.6.3
+tiled_swiglu,liger_tiled,backward,speed,ms,T,sequence length,4096,2.2056961059570312,2.1886465072631838,2.221824073791504,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:27,0.6.3
+tiled_swiglu,liger_tiled,backward,speed,ms,T,sequence length,8192,4.0389440059661865,4.035699176788331,4.042188835144043,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:27,0.6.3
+tiled_swiglu,liger_tiled,backward,speed,ms,T,sequence length,16384,8.060064315795898,8.060064315795898,8.060064315795898,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:27,0.6.3
+tiled_swiglu,deepspeed_tiled,backward,speed,ms,T,sequence length,1024,1.4678080081939697,1.449452829360962,1.5007680177688598,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:28,0.6.3
+tiled_swiglu,deepspeed_tiled,backward,speed,ms,T,sequence length,2048,1.4094560146331787,1.4087680578231812,1.4117439985275269,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:28,0.6.3
+tiled_swiglu,deepspeed_tiled,backward,speed,ms,T,sequence length,4096,2.3513920307159424,2.3492928028106688,2.3518784046173096,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:28,0.6.3
+tiled_swiglu,deepspeed_tiled,backward,speed,ms,T,sequence length,8192,4.382848024368286,4.354393577575684,4.411302471160889,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:28,0.6.3
+tiled_swiglu,deepspeed_tiled,backward,speed,ms,T,sequence length,16384,8.098272323608398,8.098272323608398,8.098272323608398,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:28,0.6.3
+tiled_swiglu,huggingface,full,memory,MB,T,sequence length,1024,312.0,312.0,312.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:29,0.6.3
+tiled_swiglu,huggingface,full,memory,MB,T,sequence length,2048,448.0,448.0,448.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:29,0.6.3
+tiled_swiglu,huggingface,full,memory,MB,T,sequence length,4096,736.0,736.0,736.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:29,0.6.3
+tiled_swiglu,huggingface,full,memory,MB,T,sequence length,8192,1312.0,1312.0,1312.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:29,0.6.3
+tiled_swiglu,huggingface,full,memory,MB,T,sequence length,16384,2464.0,2464.0,2464.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:29,0.6.3
+tiled_swiglu,liger,full,memory,MB,T,sequence length,1024,280.0,280.0,280.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:29,0.6.3
+tiled_swiglu,liger,full,memory,MB,T,sequence length,2048,384.0,384.0,384.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:29,0.6.3
+tiled_swiglu,liger,full,memory,MB,T,sequence length,4096,592.0,592.0,592.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:29,0.6.3
+tiled_swiglu,liger,full,memory,MB,T,sequence length,8192,1008.0,1008.0,1008.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:29,0.6.3
+tiled_swiglu,liger,full,memory,MB,T,sequence length,16384,1840.0,1840.0,1840.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:29,0.6.3
+tiled_swiglu,liger_tiled,full,memory,MB,T,sequence length,1024,234.0,234.0,234.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:30,0.6.3
+tiled_swiglu,liger_tiled,full,memory,MB,T,sequence length,2048,292.0,292.0,292.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:30,0.6.3
+tiled_swiglu,liger_tiled,full,memory,MB,T,sequence length,4096,408.0,408.0,408.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:30,0.6.3
+tiled_swiglu,liger_tiled,full,memory,MB,T,sequence length,8192,640.0,640.0,640.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:30,0.6.3
+tiled_swiglu,liger_tiled,full,memory,MB,T,sequence length,16384,1104.0,1104.0,1104.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:30,0.6.3
+tiled_swiglu,deepspeed_tiled,full,memory,MB,T,sequence length,1024,238.0,238.0,238.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:31,0.6.3
+tiled_swiglu,deepspeed_tiled,full,memory,MB,T,sequence length,2048,300.0,300.0,300.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:31,0.6.3
+tiled_swiglu,deepspeed_tiled,full,memory,MB,T,sequence length,4096,424.0,424.0,424.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:31,0.6.3
+tiled_swiglu,deepspeed_tiled,full,memory,MB,T,sequence length,8192,688.0,688.0,688.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:31,0.6.3
+tiled_swiglu,deepspeed_tiled,full,memory,MB,T,sequence length,16384,1216.0,1216.0,1216.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:31,0.6.3
+tiled_swiglu,huggingface,forward,memory,MB,T,sequence length,1024,192.0,192.0,192.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:31,0.6.3
+tiled_swiglu,huggingface,forward,memory,MB,T,sequence length,2048,272.0,272.0,272.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:31,0.6.3
+tiled_swiglu,huggingface,forward,memory,MB,T,sequence length,4096,432.0,432.0,432.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:31,0.6.3
+tiled_swiglu,huggingface,forward,memory,MB,T,sequence length,8192,752.0,752.0,752.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:31,0.6.3
+tiled_swiglu,huggingface,forward,memory,MB,T,sequence length,16384,1392.0,1392.0,1392.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:31,0.6.3
+tiled_swiglu,liger,forward,memory,MB,T,sequence length,1024,176.0,176.0,176.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:32,0.6.3
+tiled_swiglu,liger,forward,memory,MB,T,sequence length,2048,240.0,240.0,240.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:32,0.6.3
+tiled_swiglu,liger,forward,memory,MB,T,sequence length,4096,368.0,368.0,368.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:32,0.6.3
+tiled_swiglu,liger,forward,memory,MB,T,sequence length,8192,624.0,624.0,624.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:32,0.6.3
+tiled_swiglu,liger,forward,memory,MB,T,sequence length,16384,1136.0,1136.0,1136.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:32,0.6.3
+tiled_swiglu,liger_tiled,forward,memory,MB,T,sequence length,1024,140.0,140.0,140.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:33,0.6.3
+tiled_swiglu,liger_tiled,forward,memory,MB,T,sequence length,2048,168.0,168.0,168.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:33,0.6.3
+tiled_swiglu,liger_tiled,forward,memory,MB,T,sequence length,4096,224.0,224.0,224.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:33,0.6.3
+tiled_swiglu,liger_tiled,forward,memory,MB,T,sequence length,8192,336.0,336.0,336.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:33,0.6.3
+tiled_swiglu,liger_tiled,forward,memory,MB,T,sequence length,16384,560.0,560.0,560.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:33,0.6.3
+tiled_swiglu,deepspeed_tiled,forward,memory,MB,T,sequence length,1024,138.0,138.0,138.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:33,0.6.3
+tiled_swiglu,deepspeed_tiled,forward,memory,MB,T,sequence length,2048,164.0,164.0,164.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:33,0.6.3
+tiled_swiglu,deepspeed_tiled,forward,memory,MB,T,sequence length,4096,216.0,216.0,216.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:33,0.6.3
+tiled_swiglu,deepspeed_tiled,forward,memory,MB,T,sequence length,8192,320.0,320.0,320.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:33,0.6.3
+tiled_swiglu,deepspeed_tiled,forward,memory,MB,T,sequence length,16384,528.0,528.0,528.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:33,0.6.3
+tiled_swiglu,huggingface,backward,memory,MB,T,sequence length,1024,312.0,312.0,312.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:34,0.6.3
+tiled_swiglu,huggingface,backward,memory,MB,T,sequence length,2048,448.0,448.0,448.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:34,0.6.3
+tiled_swiglu,huggingface,backward,memory,MB,T,sequence length,4096,736.0,736.0,736.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:34,0.6.3
+tiled_swiglu,huggingface,backward,memory,MB,T,sequence length,8192,1312.0,1312.0,1312.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:34,0.6.3
+tiled_swiglu,huggingface,backward,memory,MB,T,sequence length,16384,2464.0,2464.0,2464.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:34,0.6.3
+tiled_swiglu,liger,backward,memory,MB,T,sequence length,1024,280.0,280.0,280.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:35,0.6.3
+tiled_swiglu,liger,backward,memory,MB,T,sequence length,2048,384.0,384.0,384.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:35,0.6.3
+tiled_swiglu,liger,backward,memory,MB,T,sequence length,4096,592.0,592.0,592.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:35,0.6.3
+tiled_swiglu,liger,backward,memory,MB,T,sequence length,8192,1008.0,1008.0,1008.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:35,0.6.3
+tiled_swiglu,liger,backward,memory,MB,T,sequence length,16384,1840.0,1840.0,1840.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:35,0.6.3
+tiled_swiglu,liger_tiled,backward,memory,MB,T,sequence length,1024,234.0,234.0,234.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:35,0.6.3
+tiled_swiglu,liger_tiled,backward,memory,MB,T,sequence length,2048,292.0,292.0,292.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:35,0.6.3
+tiled_swiglu,liger_tiled,backward,memory,MB,T,sequence length,4096,408.0,408.0,408.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:35,0.6.3
+tiled_swiglu,liger_tiled,backward,memory,MB,T,sequence length,8192,640.0,640.0,640.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:35,0.6.3
+tiled_swiglu,liger_tiled,backward,memory,MB,T,sequence length,16384,1104.0,1104.0,1104.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:35,0.6.3
+tiled_swiglu,deepspeed_tiled,backward,memory,MB,T,sequence length,1024,238.0,238.0,238.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:36,0.6.3
+tiled_swiglu,deepspeed_tiled,backward,memory,MB,T,sequence length,2048,300.0,300.0,300.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:36,0.6.3
+tiled_swiglu,deepspeed_tiled,backward,memory,MB,T,sequence length,4096,424.0,424.0,424.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:36,0.6.3
+tiled_swiglu,deepspeed_tiled,backward,memory,MB,T,sequence length,8192,688.0,688.0,688.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:36,0.6.3
+tiled_swiglu,deepspeed_tiled,backward,memory,MB,T,sequence length,16384,1216.0,1216.0,1216.0,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:36,0.6.3
+tvd,liger,full,memory,MB,V,vocab size,4096,1792.0009765625,1792.0009765625,1792.0009765625,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:43,0.6.3
+tvd,liger,full,memory,MB,V,vocab size,8192,3584.0009765625,3584.0009765625,3584.0009765625,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:43,0.6.3
+tvd,liger,full,memory,MB,V,vocab size,16384,7168.0009765625,7168.0009765625,7168.0009765625,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:43,0.6.3
+tvd,liger,full,memory,MB,V,vocab size,32768,14336.0009765625,14336.0009765625,14336.0009765625,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:43,0.6.3
+tvd,liger,full,memory,MB,V,vocab size,65536,28672.0,28672.0,28672.0,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:43,0.6.3
+tvd,liger,full,memory,MB,V,vocab size,131072,57344.0,57344.0,57344.0,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:43,0.6.3
+tvd,torch,full,memory,MB,V,vocab size,4096,2048.0009765625,2048.0009765625,2048.0009765625,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:43,0.6.3
+tvd,torch,full,memory,MB,V,vocab size,8192,4096.0009765625,4096.0009765625,4096.0009765625,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:43,0.6.3
+tvd,torch,full,memory,MB,V,vocab size,16384,8192.0009765625,8192.0009765625,8192.0009765625,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:43,0.6.3
+tvd,torch,full,memory,MB,V,vocab size,32768,16384.0,16384.0,16384.0,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:43,0.6.3
+tvd,torch,full,memory,MB,V,vocab size,65536,32768.0,32768.0,32768.0,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:43,0.6.3
+tvd,torch,full,memory,MB,V,vocab size,131072,65536.0,65536.0,65536.0,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:43,0.6.3
+tvd,liger,forward,speed,ms,V,vocab size,4096,0.46000000834465027,0.4591040015220642,0.4609279930591583,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:46,0.6.3
+tvd,liger,forward,speed,ms,V,vocab size,8192,0.8940480053424835,0.8930623769760132,0.8950911879539489,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:46,0.6.3
+tvd,liger,forward,speed,ms,V,vocab size,16384,1.7630240321159363,1.7614272117614744,1.7640960216522217,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:46,0.6.3
+tvd,liger,forward,speed,ms,V,vocab size,32768,3.5101120471954346,3.508537626266479,3.5116671085357662,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:46,0.6.3
+tvd,liger,forward,speed,ms,V,vocab size,65536,7.045776128768921,7.044703960418701,7.047513484954834,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:46,0.6.3
+tvd,liger,forward,speed,ms,V,vocab size,131072,14.150464057922363,14.146944046020508,14.151808166503907,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:46,0.6.3
+tvd,torch,forward,speed,ms,V,vocab size,4096,0.7311040163040161,0.730406415462494,0.7317439913749695,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:47,0.6.3
+tvd,torch,forward,speed,ms,V,vocab size,8192,1.4287680387496948,1.4279040098190308,1.4298559427261353,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:47,0.6.3
+tvd,torch,forward,speed,ms,V,vocab size,16384,2.8175359964370728,2.8164992332458496,2.8188864231109623,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:47,0.6.3
+tvd,torch,forward,speed,ms,V,vocab size,32768,5.595327854156494,5.594534206390382,5.597164821624756,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:47,0.6.3
+tvd,torch,forward,speed,ms,V,vocab size,65536,11.178175926208496,11.177177619934081,11.179027175903322,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:47,0.6.3
+tvd,torch,forward,speed,ms,V,vocab size,131072,22.35321617126465,22.351533126831058,22.355282974243167,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:47,0.6.3
+tvd,liger,full,speed,ms,V,vocab size,4096,1.3118079900741577,1.3088256359100343,1.3147199869155883,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:48,0.6.3
+tvd,liger,full,speed,ms,V,vocab size,8192,2.526848077774048,2.5248639583587646,2.532620763778686,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:48,0.6.3
+tvd,liger,full,speed,ms,V,vocab size,16384,5.264703989028931,5.2628734588623045,5.267142295837402,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:48,0.6.3
+tvd,liger,full,speed,ms,V,vocab size,32768,10.504223823547363,10.487890815734863,10.51148796081543,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:48,0.6.3
+tvd,liger,full,speed,ms,V,vocab size,65536,21.004015922546387,20.989171218872073,21.024710845947265,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:48,0.6.3
+tvd,liger,full,speed,ms,V,vocab size,131072,42.05084800720215,42.0485237121582,42.0531723022461,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:48,0.6.3
+tvd,torch,full,speed,ms,V,vocab size,4096,2.130880117416382,2.1291840076446533,2.1317888736724853,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:49,0.6.3
+tvd,torch,full,speed,ms,V,vocab size,8192,4.196800231933594,4.194470310211182,4.197504043579102,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:49,0.6.3
+tvd,torch,full,speed,ms,V,vocab size,16384,8.643872261047363,8.642463684082031,8.646719932556152,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:49,0.6.3
+tvd,torch,full,speed,ms,V,vocab size,32768,17.25372886657715,17.253472137451173,17.257419967651366,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:49,0.6.3
+tvd,torch,full,speed,ms,V,vocab size,65536,34.54296112060547,34.53945693969727,34.54646530151367,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:49,0.6.3
+tvd,torch,full,speed,ms,V,vocab size,131072,69.18192291259766,69.18192291259766,69.18192291259766,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:49,0.6.3
+tvd,liger,backward,speed,ms,V,vocab size,4096,0.85343998670578,0.8506943821907044,0.8576831936836242,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:51,0.6.3
+tvd,liger,backward,speed,ms,V,vocab size,8192,1.6309759616851807,1.6281983613967896,1.6368255853652953,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:51,0.6.3
+tvd,liger,backward,speed,ms,V,vocab size,16384,3.4991040229797363,3.4959232330322267,3.501881647109985,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:51,0.6.3
+tvd,liger,backward,speed,ms,V,vocab size,32768,6.96345591545105,6.9605631828308105,6.975104045867919,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:51,0.6.3
+tvd,liger,backward,speed,ms,V,vocab size,65536,13.945759773254395,13.926694488525392,13.987411499023438,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:51,0.6.3
+tvd,liger,backward,speed,ms,V,vocab size,131072,27.88979148864746,27.873356628417966,27.890867233276367,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:51,0.6.3
+tvd,torch,backward,speed,ms,V,vocab size,4096,1.4023360013961792,1.4009920358657837,1.4034240245819092,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:52,0.6.3
+tvd,torch,backward,speed,ms,V,vocab size,8192,2.772208094596863,2.7709824085235595,2.7734015464782713,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:52,0.6.3
+tvd,torch,backward,speed,ms,V,vocab size,16384,5.830143928527832,5.82912015914917,5.8321919441223145,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:52,0.6.3
+tvd,torch,backward,speed,ms,V,vocab size,32768,11.662863731384277,11.662694358825684,11.663296127319336,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:52,0.6.3
+tvd,torch,backward,speed,ms,V,vocab size,65536,23.363823890686035,23.359257507324216,23.370937728881835,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:52,0.6.3
+tvd,torch,backward,speed,ms,V,vocab size,131072,46.831119537353516,46.829592895507815,46.83264617919922,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-11-21 01:01:52,0.6.3
diff --git a/benchmarks/commits.txt b/benchmarks/commits.txt
index 18d82987..85e95ddf 100644
--- a/benchmarks/commits.txt
+++ b/benchmarks/commits.txt
@@ -14,3 +14,4 @@ c856fba
 d5648bf-v0.6.3
 a80f30b
 817436f
+4c32ab6