diff --git a/benchmarks/6059dfb/benchmark.csv b/benchmarks/6059dfb/benchmark.csv
new file mode 100644
index 00000000..ad011c47
--- /dev/null
+++ b/benchmarks/6059dfb/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,2,12.253024101257324,12.053376007080079,12.709587097167969,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:13:21,0.6.4
+fused_linear_cpo_loss,liger,forward,speed,ms,B,B,4,24.29684829711914,24.227283477783203,24.41242904663086,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:13:21,0.6.4
+fused_linear_cpo_loss,liger,forward,speed,ms,B,B,8,49.0549430847168,48.7797492980957,49.33013687133789,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:13:21,0.6.4
+fused_linear_cpo_loss,liger,forward,speed,ms,B,B,16,95.86460876464844,95.86460876464844,95.86460876464844,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:13:21,0.6.4
+fused_linear_cpo_loss,huggingface,forward,speed,ms,B,B,2,6.147520065307617,6.140684700012207,6.1547393798828125,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:13:45,0.6.4
+fused_linear_cpo_loss,huggingface,forward,speed,ms,B,B,4,12.007152080535889,12.00197105407715,12.028537559509278,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:13:45,0.6.4
+fused_linear_cpo_loss,huggingface,forward,speed,ms,B,B,8,23.59235191345215,23.550496292114257,23.65929527282715,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:13:45,0.6.4
+fused_linear_cpo_loss,huggingface,forward,speed,ms,B,B,16,47.06145477294922,46.93808517456055,47.184824371337896,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:13:45,0.6.4
+fused_linear_cpo_loss,liger,backward,speed,ms,B,B,2,1.0939199924468994,1.091488003730774,1.1039040088653564,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:14:08,0.6.4
+fused_linear_cpo_loss,liger,backward,speed,ms,B,B,4,1.105023980140686,1.1022976398468018,1.1201600313186646,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:14:08,0.6.4
+fused_linear_cpo_loss,liger,backward,speed,ms,B,B,8,1.1250239610671997,1.1239871740341187,1.1285247802734375,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:14:08,0.6.4
+fused_linear_cpo_loss,liger,backward,speed,ms,B,B,16,1.1692320108413696,1.1679359674453735,1.1720319986343384,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:14:08,0.6.4
+fused_linear_cpo_loss,huggingface,backward,speed,ms,B,B,2,11.174032211303711,11.166246604919433,11.178368186950683,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:14:32,0.6.4
+fused_linear_cpo_loss,huggingface,backward,speed,ms,B,B,4,21.230223655700684,21.215686035156253,21.334246444702146,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:14:32,0.6.4
+fused_linear_cpo_loss,huggingface,backward,speed,ms,B,B,8,41.43180847167969,41.43167343139649,41.43194351196289,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:14:32,0.6.4
+fused_linear_cpo_loss,huggingface,backward,speed,ms,B,B,16,84.19110107421875,84.19110107421875,84.19110107421875,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:14:32,0.6.4
+fused_linear_cpo_loss,liger,full,speed,ms,B,B,2,13.762016296386719,13.749779319763183,13.781785202026366,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:14:56,0.6.4
+fused_linear_cpo_loss,liger,full,speed,ms,B,B,4,25.87161636352539,25.67848358154297,25.973875045776367,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:14:56,0.6.4
+fused_linear_cpo_loss,liger,full,speed,ms,B,B,8,49.941375732421875,49.941375732421875,49.941375732421875,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:14:56,0.6.4
+fused_linear_cpo_loss,liger,full,speed,ms,B,B,16,99.82361602783203,99.82361602783203,99.82361602783203,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:14:56,0.6.4
+fused_linear_cpo_loss,huggingface,full,speed,ms,B,B,2,17.3984317779541,17.39423370361328,17.4139835357666,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:15:21,0.6.4
+fused_linear_cpo_loss,huggingface,full,speed,ms,B,B,4,33.41297721862793,33.41137390136719,33.414580535888675,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:15:21,0.6.4
+fused_linear_cpo_loss,huggingface,full,speed,ms,B,B,8,66.34473419189453,66.34473419189453,66.34473419189453,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:15:21,0.6.4
+fused_linear_cpo_loss,huggingface,full,speed,ms,B,B,16,134.86671447753906,134.86671447753906,134.86671447753906,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:15:21,0.6.4
+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 HBM3,2025-12-19 00:15:45,0.6.4
+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 HBM3,2025-12-19 00:15:45,0.6.4
+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 HBM3,2025-12-19 00:15:45,0.6.4
+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 HBM3,2025-12-19 00:15:45,0.6.4
+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 HBM3,2025-12-19 00:16:10,0.6.4
+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 HBM3,2025-12-19 00:16:10,0.6.4
+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 HBM3,2025-12-19 00:16:10,0.6.4
+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 HBM3,2025-12-19 00:16:10,0.6.4
+cross_entropy,liger,forward,speed,ms,V,vocab size,4096,0.4036479890346527,0.4007743954658508,0.40711040496826173,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:18,0.6.4
+cross_entropy,liger,forward,speed,ms,V,vocab size,8192,0.532863974571228,0.530124819278717,0.5349760055541992,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:18,0.6.4
+cross_entropy,liger,forward,speed,ms,V,vocab size,16384,1.017296016216278,1.0134719610214233,1.0212479829788208,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:18,0.6.4
+cross_entropy,liger,forward,speed,ms,V,vocab size,32768,1.658944010734558,1.656620764732361,1.6620095729827882,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:18,0.6.4
+cross_entropy,liger,forward,speed,ms,V,vocab size,65536,4.079008102416992,4.068883037567139,4.084140872955322,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:18,0.6.4
+cross_entropy,liger,forward,speed,ms,V,vocab size,131072,8.775232315063477,8.769503593444824,8.782431602478027,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:18,0.6.4
+cross_entropy,huggingface,forward,speed,ms,V,vocab size,4096,0.24451200664043427,0.24377599358558655,0.24545920193195342,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:19,0.6.4
+cross_entropy,huggingface,forward,speed,ms,V,vocab size,8192,0.5492479801177979,0.5472575783729553,0.5510335922241211,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:19,0.6.4
+cross_entropy,huggingface,forward,speed,ms,V,vocab size,16384,0.8065280020236969,0.8055999875068665,0.8074560165405273,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:19,0.6.4
+cross_entropy,huggingface,forward,speed,ms,V,vocab size,32768,2.711359977722168,2.7105408668518067,2.7131200313568113,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:19,0.6.4
+cross_entropy,huggingface,forward,speed,ms,V,vocab size,65536,5.593984127044678,5.593088150024414,5.595481586456298,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:19,0.6.4
+cross_entropy,huggingface,forward,speed,ms,V,vocab size,131072,11.302752017974854,11.295616340637206,11.318015861511231,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:19,0.6.4
+cross_entropy,liger,backward,speed,ms,V,vocab size,4096,0.2324800044298172,0.23162880539894104,0.23297280371189116,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:20,0.6.4
+cross_entropy,liger,backward,speed,ms,V,vocab size,8192,0.4121119976043701,0.4111552059650421,0.415014386177063,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:20,0.6.4
+cross_entropy,liger,backward,speed,ms,V,vocab size,16384,0.76528000831604,0.7642367959022522,0.7674048066139221,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:20,0.6.4
+cross_entropy,liger,backward,speed,ms,V,vocab size,32768,1.4721439480781555,1.4706111907958985,1.473676824569702,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:20,0.6.4
+cross_entropy,liger,backward,speed,ms,V,vocab size,65536,2.8901920318603516,2.887750482559204,2.8921854972839354,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:20,0.6.4
+cross_entropy,liger,backward,speed,ms,V,vocab size,131072,5.735712051391602,5.731296062469482,5.738368034362793,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:20,0.6.4
+cross_entropy,huggingface,backward,speed,ms,V,vocab size,4096,0.376800000667572,0.3758783876895904,0.3784320116043091,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:20,0.6.4
+cross_entropy,huggingface,backward,speed,ms,V,vocab size,8192,0.7224799990653992,0.720851194858551,0.7239808082580566,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:20,0.6.4
+cross_entropy,huggingface,backward,speed,ms,V,vocab size,16384,1.605504035949707,1.604640007019043,1.6067520380020142,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:20,0.6.4
+cross_entropy,huggingface,backward,speed,ms,V,vocab size,32768,3.5261759757995605,3.5247808456420895,3.527206468582153,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:20,0.6.4
+cross_entropy,huggingface,backward,speed,ms,V,vocab size,65536,7.064079999923706,7.0613631248474125,7.067039966583252,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:20,0.6.4
+cross_entropy,huggingface,backward,speed,ms,V,vocab size,131072,14.127344131469727,14.12282886505127,14.138361549377443,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:20,0.6.4
+cross_entropy,liger,full,speed,ms,V,vocab size,4096,0.6587200164794922,0.65137277841568,0.6712128162384033,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:21,0.6.4
+cross_entropy,liger,full,speed,ms,V,vocab size,8192,0.9661120176315308,0.9595199823379517,0.9775359988212585,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:21,0.6.4
+cross_entropy,liger,full,speed,ms,V,vocab size,16384,1.807744026184082,1.8036799430847168,1.8257919549942017,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:21,0.6.4
+cross_entropy,liger,full,speed,ms,V,vocab size,32768,3.164639949798584,3.1605631828308103,3.169542407989502,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:21,0.6.4
+cross_entropy,liger,full,speed,ms,V,vocab size,65536,7.014863967895508,7.004787349700928,7.016876602172852,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:21,0.6.4
+cross_entropy,liger,full,speed,ms,V,vocab size,131072,14.557744026184082,14.556511878967285,14.561056137084961,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:21,0.6.4
+cross_entropy,huggingface,full,speed,ms,V,vocab size,4096,0.8845440149307251,0.8836927890777588,0.8858879804611206,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:22,0.6.4
+cross_entropy,huggingface,full,speed,ms,V,vocab size,8192,1.7955199480056763,1.792844820022583,1.797926449775696,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:22,0.6.4
+cross_entropy,huggingface,full,speed,ms,V,vocab size,16384,3.4617120027542114,3.46014084815979,3.463327932357788,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:22,0.6.4
+cross_entropy,huggingface,full,speed,ms,V,vocab size,32768,8.343135833740234,8.341631889343262,8.344639778137207,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:22,0.6.4
+cross_entropy,huggingface,full,speed,ms,V,vocab size,65536,16.868864059448242,16.867673110961913,16.87413673400879,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:22,0.6.4
+cross_entropy,huggingface,full,speed,ms,V,vocab size,131072,33.86508750915527,33.86030731201172,33.86986770629883,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:22,0.6.4
+cross_entropy,liger,no-grad-forward,speed,ms,V,vocab size,4096,0.4020320028066635,0.39955198764801025,0.4052159905433655,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:23,0.6.4
+cross_entropy,liger,no-grad-forward,speed,ms,V,vocab size,8192,0.5316159725189209,0.529151976108551,0.5346879959106445,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:23,0.6.4
+cross_entropy,liger,no-grad-forward,speed,ms,V,vocab size,16384,1.0155200362205505,1.0119552135467529,1.0187328338623047,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:23,0.6.4
+cross_entropy,liger,no-grad-forward,speed,ms,V,vocab size,32768,1.6570240259170532,1.654310417175293,1.6606399774551392,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:23,0.6.4
+cross_entropy,liger,no-grad-forward,speed,ms,V,vocab size,65536,4.072784185409546,4.068537712097168,4.077740955352784,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:23,0.6.4
+cross_entropy,liger,no-grad-forward,speed,ms,V,vocab size,131072,8.765664100646973,8.761216163635254,8.769536018371582,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:23,0.6.4
+cross_entropy,huggingface,no-grad-forward,speed,ms,V,vocab size,4096,0.24463999271392822,0.24380800127983093,0.24537600576877594,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:24,0.6.4
+cross_entropy,huggingface,no-grad-forward,speed,ms,V,vocab size,8192,0.5495039820671082,0.547763180732727,0.5510976076126098,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:24,0.6.4
+cross_entropy,huggingface,no-grad-forward,speed,ms,V,vocab size,16384,0.8063199818134308,0.805510413646698,0.807200014591217,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:24,0.6.4
+cross_entropy,huggingface,no-grad-forward,speed,ms,V,vocab size,32768,2.711616039276123,2.7092160224914554,2.7126975536346434,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:24,0.6.4
+cross_entropy,huggingface,no-grad-forward,speed,ms,V,vocab size,65536,5.594207763671875,5.59290246963501,5.595961475372315,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:24,0.6.4
+cross_entropy,huggingface,no-grad-forward,speed,ms,V,vocab size,131072,11.299647808074951,11.29184627532959,11.305491065979005,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:24,0.6.4
+cross_entropy,liger,full,memory,MB,V,vocab size,4096,256.32861328125,256.32861328125,256.32861328125,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:24,0.6.4
+cross_entropy,liger,full,memory,MB,V,vocab size,8192,512.32861328125,512.32861328125,512.32861328125,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:24,0.6.4
+cross_entropy,liger,full,memory,MB,V,vocab size,16384,1024.32861328125,1024.32861328125,1024.32861328125,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:24,0.6.4
+cross_entropy,liger,full,memory,MB,V,vocab size,32768,2048.32861328125,2048.32861328125,2048.32861328125,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:24,0.6.4
+cross_entropy,liger,full,memory,MB,V,vocab size,65536,4096.32861328125,4096.32861328125,4096.32861328125,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:24,0.6.4
+cross_entropy,liger,full,memory,MB,V,vocab size,131072,8192.328125,8192.328125,8192.328125,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:24,0.6.4
+cross_entropy,huggingface,full,memory,MB,V,vocab size,4096,1280.1259765625,1280.1259765625,1280.1259765625,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:24,0.6.4
+cross_entropy,huggingface,full,memory,MB,V,vocab size,8192,2560.1259765625,2560.1259765625,2560.1259765625,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:24,0.6.4
+cross_entropy,huggingface,full,memory,MB,V,vocab size,16384,5120.1259765625,5120.1259765625,5120.1259765625,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:24,0.6.4
+cross_entropy,huggingface,full,memory,MB,V,vocab size,32768,10240.1259765625,10240.1259765625,10240.1259765625,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:24,0.6.4
+cross_entropy,huggingface,full,memory,MB,V,vocab size,65536,20480.125,20480.125,20480.125,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:24,0.6.4
+cross_entropy,huggingface,full,memory,MB,V,vocab size,131072,40960.125,40960.125,40960.125,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:16:24,0.6.4
+distill_cosine_loss,liger,forward,speed,ms,BT,B x T,1024,5.56766414642334,5.509759902954102,5.577311992645264,"{""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 HBM3,2025-12-19 00:17:17,0.6.4
+distill_cosine_loss,liger,forward,speed,ms,BT,B x T,2048,11.273712158203125,11.001747322082519,11.341471862792968,"{""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 HBM3,2025-12-19 00:17:17,0.6.4
+distill_cosine_loss,liger,forward,speed,ms,BT,B x T,4096,22.552032470703125,22.48776969909668,22.627225875854492,"{""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 HBM3,2025-12-19 00:17:17,0.6.4
+distill_cosine_loss,liger,forward,speed,ms,BT,B x T,8192,45.78635215759277,45.624506378173834,45.94819793701172,"{""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 HBM3,2025-12-19 00:17:17,0.6.4
+distill_cosine_loss,torch,forward,speed,ms,BT,B x T,1024,7.588495969772339,7.58452491760254,7.590585708618164,"{""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 HBM3,2025-12-19 00:17:52,0.6.4
+distill_cosine_loss,torch,forward,speed,ms,BT,B x T,2048,14.845727920532227,14.83737564086914,14.851872444152832,"{""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 HBM3,2025-12-19 00:17:52,0.6.4
+distill_cosine_loss,torch,forward,speed,ms,BT,B x T,4096,29.303808212280273,29.28326377868652,29.309357452392582,"{""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 HBM3,2025-12-19 00:17:52,0.6.4
+distill_cosine_loss,torch,forward,speed,ms,BT,B x T,8192,59.20051193237305,59.20051193237305,59.20051193237305,"{""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 HBM3,2025-12-19 00:17:52,0.6.4
+distill_cosine_loss,liger,full,speed,ms,BT,B x T,1024,7.4007041454315186,7.390963363647461,7.450879859924317,"{""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 HBM3,2025-12-19 00:18:27,0.6.4
+distill_cosine_loss,liger,full,speed,ms,BT,B x T,2048,12.834015846252441,12.825286102294921,12.88327693939209,"{""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 HBM3,2025-12-19 00:18:27,0.6.4
+distill_cosine_loss,liger,full,speed,ms,BT,B x T,4096,23.981696128845215,23.94357795715332,24.002879333496093,"{""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 HBM3,2025-12-19 00:18:27,0.6.4
+distill_cosine_loss,liger,full,speed,ms,BT,B x T,8192,46.44163131713867,46.323551177978516,46.55971145629883,"{""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 HBM3,2025-12-19 00:18:27,0.6.4
+distill_cosine_loss,torch,full,speed,ms,BT,B x T,1024,20.339248657226562,20.328358840942386,20.350368118286134,"{""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 HBM3,2025-12-19 00:19:03,0.6.4
+distill_cosine_loss,torch,full,speed,ms,BT,B x T,2048,39.47983932495117,39.47066116333008,39.489017486572266,"{""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 HBM3,2025-12-19 00:19:03,0.6.4
+distill_cosine_loss,torch,full,speed,ms,BT,B x T,4096,78.48560333251953,78.48560333251953,78.48560333251953,"{""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 HBM3,2025-12-19 00:19:03,0.6.4
+distill_cosine_loss,torch,full,speed,ms,BT,B x T,8192,156.80770874023438,156.80770874023438,156.80770874023438,"{""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 HBM3,2025-12-19 00:19:03,0.6.4
+distill_cosine_loss,liger,full,memory,MB,BT,B x T,1024,5107.0185546875,5107.0185546875,5107.0185546875,"{""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 HBM3,2025-12-19 00:19:37,0.6.4
+distill_cosine_loss,liger,full,memory,MB,BT,B x T,2048,5135.0263671875,5135.0263671875,5135.0263671875,"{""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 HBM3,2025-12-19 00:19:37,0.6.4
+distill_cosine_loss,liger,full,memory,MB,BT,B x T,4096,5191.0419921875,5191.0419921875,5191.0419921875,"{""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 HBM3,2025-12-19 00:19:37,0.6.4
+distill_cosine_loss,liger,full,memory,MB,BT,B x T,8192,5303.0732421875,5303.0732421875,5303.0732421875,"{""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 HBM3,2025-12-19 00:19:37,0.6.4
+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 HBM3,2025-12-19 00:20:13,0.6.4
+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 HBM3,2025-12-19 00:20:13,0.6.4
+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 HBM3,2025-12-19 00:20:13,0.6.4
+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 HBM3,2025-12-19 00:20:13,0.6.4
+distill_jsd_loss,liger,forward,speed,ms,BT,B x T,1024,7.435712099075317,7.407532691955566,7.465631866455078,"{""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 HBM3,2025-12-19 00:21:09,0.6.4
+distill_jsd_loss,liger,forward,speed,ms,BT,B x T,2048,14.920432090759277,14.811200141906738,14.925919532775879,"{""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 HBM3,2025-12-19 00:21:09,0.6.4
+distill_jsd_loss,liger,forward,speed,ms,BT,B x T,4096,29.78883171081543,29.779558563232424,29.926553344726564,"{""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 HBM3,2025-12-19 00:21:09,0.6.4
+distill_jsd_loss,liger,forward,speed,ms,BT,B x T,8192,59.241310119628906,59.241310119628906,59.241310119628906,"{""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 HBM3,2025-12-19 00:21:09,0.6.4
+distill_jsd_loss,torch,forward,speed,ms,BT,B x T,1024,16.38115119934082,16.37001609802246,16.38502311706543,"{""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 HBM3,2025-12-19 00:21:45,0.6.4
+distill_jsd_loss,torch,forward,speed,ms,BT,B x T,2048,32.37209701538086,32.36835250854492,32.391508483886724,"{""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 HBM3,2025-12-19 00:21:45,0.6.4
+distill_jsd_loss,torch,forward,speed,ms,BT,B x T,4096,67.75836944580078,67.75836944580078,67.75836944580078,"{""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 HBM3,2025-12-19 00:21:45,0.6.4
+distill_jsd_loss,torch,forward,speed,ms,BT,B x T,8192,135.30592346191406,135.30592346191406,135.30592346191406,"{""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 HBM3,2025-12-19 00:21:45,0.6.4
+distill_jsd_loss,liger,backward,speed,ms,BT,B x T,1024,0.5722560286521912,0.571008026599884,0.5762879848480225,"{""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 HBM3,2025-12-19 00:22:20,0.6.4
+distill_jsd_loss,liger,backward,speed,ms,BT,B x T,2048,0.5732799768447876,0.5716800093650818,0.575872004032135,"{""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 HBM3,2025-12-19 00:22:20,0.6.4
+distill_jsd_loss,liger,backward,speed,ms,BT,B x T,4096,0.5772479772567749,0.576307189464569,0.5789120197296143,"{""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 HBM3,2025-12-19 00:22:20,0.6.4
+distill_jsd_loss,liger,backward,speed,ms,BT,B x T,8192,0.5887680053710938,0.587603223323822,0.5903488039970398,"{""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 HBM3,2025-12-19 00:22:20,0.6.4
+distill_jsd_loss,torch,backward,speed,ms,BT,B x T,1024,12.51910400390625,12.517772483825684,12.521478652954102,"{""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 HBM3,2025-12-19 00:22:56,0.6.4
+distill_jsd_loss,torch,backward,speed,ms,BT,B x T,2048,24.2925443649292,24.289574432373044,24.29593620300293,"{""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 HBM3,2025-12-19 00:22:56,0.6.4
+distill_jsd_loss,torch,backward,speed,ms,BT,B x T,4096,47.16217613220215,47.16190719604493,47.162445068359375,"{""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 HBM3,2025-12-19 00:22:56,0.6.4
+distill_jsd_loss,torch,backward,speed,ms,BT,B x T,8192,93.90656280517578,93.90656280517578,93.90656280517578,"{""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 HBM3,2025-12-19 00:22:56,0.6.4
+distill_jsd_loss,liger,full,speed,ms,BT,B x T,1024,9.364367961883545,9.331404495239257,9.405606460571288,"{""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 HBM3,2025-12-19 00:23:31,0.6.4
+distill_jsd_loss,liger,full,speed,ms,BT,B x T,2048,16.708383560180664,16.655059814453125,16.73509101867676,"{""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 HBM3,2025-12-19 00:23:31,0.6.4
+distill_jsd_loss,liger,full,speed,ms,BT,B x T,4096,31.869792938232422,31.759660720825195,32.059316253662104,"{""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 HBM3,2025-12-19 00:23:31,0.6.4
+distill_jsd_loss,liger,full,speed,ms,BT,B x T,8192,61.42108917236328,61.42108917236328,61.42108917236328,"{""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 HBM3,2025-12-19 00:23:31,0.6.4
+distill_jsd_loss,torch,full,speed,ms,BT,B x T,1024,28.91004753112793,28.906687545776368,28.913138580322265,"{""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 HBM3,2025-12-19 00:24:08,0.6.4
+distill_jsd_loss,torch,full,speed,ms,BT,B x T,2048,56.73548889160156,56.73548889160156,56.73548889160156,"{""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 HBM3,2025-12-19 00:24:08,0.6.4
+distill_jsd_loss,torch,full,speed,ms,BT,B x T,4096,115.00012969970703,115.00012969970703,115.00012969970703,"{""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 HBM3,2025-12-19 00:24:08,0.6.4
+distill_jsd_loss,torch,full,speed,ms,BT,B x T,8192,229.42807006835938,229.42807006835938,229.42807006835938,"{""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 HBM3,2025-12-19 00:24:08,0.6.4
+distill_jsd_loss,liger,full,memory,MB,BT,B x T,1024,6102.0341796875,6102.0341796875,6102.0341796875,"{""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 HBM3,2025-12-19 00:24:43,0.6.4
+distill_jsd_loss,liger,full,memory,MB,BT,B x T,2048,6130.0419921875,6130.0419921875,6130.0419921875,"{""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 HBM3,2025-12-19 00:24:43,0.6.4
+distill_jsd_loss,liger,full,memory,MB,BT,B x T,4096,6186.0576171875,6186.0576171875,6186.0576171875,"{""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 HBM3,2025-12-19 00:24:43,0.6.4
+distill_jsd_loss,liger,full,memory,MB,BT,B x T,8192,6298.0888671875,6298.0888671875,6298.0888671875,"{""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 HBM3,2025-12-19 00:24:43,0.6.4
+distill_jsd_loss,torch,full,memory,MB,BT,B x T,1024,11613.0087890625,11613.0087890625,11613.0087890625,"{""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 HBM3,2025-12-19 00:25:21,0.6.4
+distill_jsd_loss,torch,full,memory,MB,BT,B x T,2048,19653.015625,19653.015625,19653.015625,"{""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 HBM3,2025-12-19 00:25:21,0.6.4
+distill_jsd_loss,torch,full,memory,MB,BT,B x T,4096,35733.03125,35733.03125,35733.03125,"{""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 HBM3,2025-12-19 00:25:21,0.6.4
+distill_jsd_loss,torch,full,memory,MB,BT,B x T,8192,67893.0625,67893.0625,67893.0625,"{""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 HBM3,2025-12-19 00:25:21,0.6.4
+dpo_loss,liger,forward,speed,ms,B,Batch Size (B),2,3.504512071609497,3.50130558013916,3.519487953186035,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:25:55,0.6.4
+dpo_loss,liger,forward,speed,ms,B,Batch Size (B),4,6.910975933074951,6.905734443664551,6.916441535949707,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:25:55,0.6.4
+dpo_loss,liger,forward,speed,ms,B,Batch Size (B),8,13.565216064453125,13.55779857635498,13.624595451354981,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:25:55,0.6.4
+dpo_loss,liger,forward,speed,ms,B,Batch Size (B),16,27.36751937866211,27.366770935058593,27.3728385925293,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:25:55,0.6.4
+dpo_loss,liger,forward,speed,ms,B,Batch Size (B),32,54.476863861083984,54.476863861083984,54.476863861083984,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:25:55,0.6.4
+dpo_loss,huggingface,forward,speed,ms,B,Batch Size (B),2,2.918463945388794,2.9169407367706297,2.920659112930298,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:26:10,0.6.4
+dpo_loss,huggingface,forward,speed,ms,B,Batch Size (B),4,5.5465922355651855,5.540774536132813,5.561823940277099,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:26:10,0.6.4
+dpo_loss,huggingface,forward,speed,ms,B,Batch Size (B),8,10.80339241027832,10.795065689086915,10.823775672912596,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:26:10,0.6.4
+dpo_loss,huggingface,forward,speed,ms,B,Batch Size (B),16,21.202320098876953,21.199788284301757,21.203622817993164,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:26:10,0.6.4
+dpo_loss,huggingface,forward,speed,ms,B,Batch Size (B),32,42.27172660827637,42.265227508544925,42.27822570800782,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:26:10,0.6.4
+dpo_loss,liger,backward,speed,ms,B,Batch Size (B),2,0.3092000037431717,0.3076159954071045,0.31174400448799133,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:26:25,0.6.4
+dpo_loss,liger,backward,speed,ms,B,Batch Size (B),4,0.3088639974594116,0.30768001079559326,0.3104639947414398,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:26:25,0.6.4
+dpo_loss,liger,backward,speed,ms,B,Batch Size (B),8,0.30953601002693176,0.30819201469421387,0.31094399094581604,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:26:25,0.6.4
+dpo_loss,liger,backward,speed,ms,B,Batch Size (B),16,0.3087039887905121,0.30748799443244934,0.3097088038921356,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:26:25,0.6.4
+dpo_loss,liger,backward,speed,ms,B,Batch Size (B),32,0.30880001187324524,0.30768001079559326,0.3102400004863739,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:26:25,0.6.4
+dpo_loss,huggingface,backward,speed,ms,B,Batch Size (B),2,1.9241759777069092,1.922214412689209,1.9253952264785767,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:26:40,0.6.4
+dpo_loss,huggingface,backward,speed,ms,B,Batch Size (B),4,3.449984073638916,3.447782373428345,3.451904058456421,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:26:40,0.6.4
+dpo_loss,huggingface,backward,speed,ms,B,Batch Size (B),8,6.482687950134277,6.4797438621521,6.4881792068481445,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:26:40,0.6.4
+dpo_loss,huggingface,backward,speed,ms,B,Batch Size (B),16,12.558112144470215,12.55712013244629,12.562348365783691,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:26:40,0.6.4
+dpo_loss,huggingface,backward,speed,ms,B,Batch Size (B),32,24.55128002166748,24.547251510620114,24.558624267578125,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:26:40,0.6.4
+dpo_loss,liger,full,speed,ms,B,Batch Size (B),2,5.629792213439941,5.621919918060303,5.640646362304688,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:26:55,0.6.4
+dpo_loss,liger,full,speed,ms,B,Batch Size (B),4,8.996031761169434,8.9474308013916,9.015693092346192,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:26:55,0.6.4
+dpo_loss,liger,full,speed,ms,B,Batch Size (B),8,15.90060806274414,15.873279571533203,15.938528060913086,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:26:55,0.6.4
+dpo_loss,liger,full,speed,ms,B,Batch Size (B),16,29.530784606933594,29.51296615600586,29.538502502441407,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:26:55,0.6.4
+dpo_loss,liger,full,speed,ms,B,Batch Size (B),32,56.61561584472656,56.61561584472656,56.61561584472656,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:26:55,0.6.4
+dpo_loss,huggingface,full,speed,ms,B,Batch Size (B),2,4.844799995422363,4.835801696777343,4.849926280975342,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:27:11,0.6.4
+dpo_loss,huggingface,full,speed,ms,B,Batch Size (B),4,9.016848087310791,9.012288284301757,9.025983810424805,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:27:11,0.6.4
+dpo_loss,huggingface,full,speed,ms,B,Batch Size (B),8,17.335039138793945,17.30321922302246,17.356691360473633,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:27:11,0.6.4
+dpo_loss,huggingface,full,speed,ms,B,Batch Size (B),16,33.76751899719238,33.76429290771485,33.77074508666992,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:27:11,0.6.4
+dpo_loss,huggingface,full,speed,ms,B,Batch Size (B),32,67.07584381103516,67.07584381103516,67.07584381103516,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:27:11,0.6.4
+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 HBM3,2025-12-19 00:27:26,0.6.4
+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 HBM3,2025-12-19 00:27:26,0.6.4
+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 HBM3,2025-12-19 00:27:26,0.6.4
+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 HBM3,2025-12-19 00:27:26,0.6.4
+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 HBM3,2025-12-19 00:27:26,0.6.4
+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 HBM3,2025-12-19 00:27:42,0.6.4
+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 HBM3,2025-12-19 00:27:42,0.6.4
+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 HBM3,2025-12-19 00:27:42,0.6.4
+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 HBM3,2025-12-19 00:27:42,0.6.4
+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 HBM3,2025-12-19 00:27:42,0.6.4
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,1024,0.012160000391304493,0.011935999616980553,0.012384000234305859,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:00,0.6.4
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,2048,0.01788800023496151,0.017632000148296356,0.01817600056529045,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:00,0.6.4
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,3072,0.023391999304294586,0.023104000836610794,0.02380800060927868,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:00,0.6.4
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,4096,0.02956799976527691,0.02924799919128418,0.030239999294281006,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:00,0.6.4
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,5120,0.03699199855327606,0.03683200106024742,0.03776000067591667,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:00,0.6.4
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,6144,0.04089599847793579,0.04070400074124336,0.04185599833726883,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:00,0.6.4
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,7168,0.048128001391887665,0.04790399968624115,0.048895999789237976,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:00,0.6.4
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,8192,0.05180799961090088,0.05142400041222572,0.05270399898290634,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:00,0.6.4
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,9216,0.05951999872922897,0.058880001306533813,0.06006399914622307,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:00,0.6.4
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,10240,0.06323199719190598,0.06239999830722809,0.06371200084686279,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:00,0.6.4
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,11264,0.07075200229883194,0.06995200365781784,0.0711359977722168,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:00,0.6.4
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,12288,0.07472000271081924,0.07369600236415863,0.07500799745321274,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:00,0.6.4
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,13312,0.0817599967122078,0.08086399734020233,0.08204799890518188,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:00,0.6.4
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,14336,0.08556800335645676,0.08489599823951721,0.08582399785518646,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:00,0.6.4
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,15360,0.09296000003814697,0.09257599711418152,0.09324800223112106,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:00,0.6.4
+dyt_beta=False,liger,forward,speed,ms,hidden_size,hidden_size,16384,0.09663999825716019,0.0963520035147667,0.0968639999628067,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:00,0.6.4
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,1024,0.04806400090456009,0.04771199822425842,0.04851200059056282,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:08,0.6.4
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,2048,0.0963520035147667,0.09593600034713745,0.09667199850082397,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:08,0.6.4
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,3072,0.1387840062379837,0.13833600282669067,0.13939200341701508,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:08,0.6.4
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,4096,0.17689600586891174,0.17638400197029114,0.177279993891716,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:08,0.6.4
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,5120,0.21740800142288208,0.2170879989862442,0.21779200434684753,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:08,0.6.4
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,6144,0.25785601139068604,0.25705599784851074,0.258432000875473,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:08,0.6.4
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,7168,0.2982720136642456,0.29791998863220215,0.2986240088939667,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:08,0.6.4
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,8192,0.34068799018859863,0.34031999111175537,0.3412800133228302,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:08,0.6.4
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,9216,0.38124799728393555,0.3807680010795593,0.3816959857940674,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:08,0.6.4
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,10240,0.4190399944782257,0.41865599155426025,0.41945600509643555,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:08,0.6.4
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,11264,0.4596160054206848,0.45903998613357544,0.4601919949054718,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:08,0.6.4
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,12288,0.5023360252380371,0.5019199848175049,0.5027520060539246,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:08,0.6.4
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,13312,0.5401920080184937,0.5397120118141174,0.5409536123275757,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:08,0.6.4
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,14336,0.5832639932632446,0.5825279951095581,0.5837439894676208,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:08,0.6.4
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,15360,0.6232640147209167,0.6228479743003845,0.623744010925293,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:08,0.6.4
+dyt_beta=False,torch,forward,speed,ms,hidden_size,hidden_size,16384,0.6619200110435486,0.661408007144928,0.6624320149421692,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:08,0.6.4
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,1024,0.03283200040459633,0.03171199932694435,0.034304000437259674,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:22,0.6.4
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,2048,0.03270399942994118,0.03159039840102196,0.03417599946260452,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:22,0.6.4
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,3072,0.035999998450279236,0.03519999980926514,0.03699199855327606,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:22,0.6.4
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,4096,0.03996799886226654,0.039583999663591385,0.040863998234272,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:22,0.6.4
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,5120,0.048448000103235245,0.0480320006608963,0.049375999718904495,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:22,0.6.4
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,6144,0.05708799883723259,0.056384000927209854,0.057792000472545624,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:22,0.6.4
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,7168,0.0658240020275116,0.06486400216817856,0.06627199798822403,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:22,0.6.4
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,8192,0.07420799881219864,0.07318399846553802,0.07459200173616409,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:22,0.6.4
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,9216,0.3815999925136566,0.381056010723114,0.38198399543762207,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:22,0.6.4
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,10240,0.4217279851436615,0.42134401202201843,0.4222080111503601,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:22,0.6.4
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,11264,0.4619840085506439,0.4613119959831238,0.4625279903411865,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:22,0.6.4
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,12288,0.49955201148986816,0.4991999864578247,0.49987199902534485,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:22,0.6.4
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,13312,0.5426239967346191,0.5421440005302429,0.5433599948883057,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:22,0.6.4
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,14336,0.5832319855690002,0.5825600028038025,0.5837439894676208,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:22,0.6.4
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,15360,0.6212159991264343,0.6207680106163025,0.6217280030250549,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:22,0.6.4
+dyt_beta=False,torch_compile,forward,speed,ms,hidden_size,hidden_size,16384,0.6618559956550598,0.6612480282783508,0.6624000072479248,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:22,0.6.4
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,1024,0.1518079936504364,0.14401919841766359,0.15695999562740326,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:31,0.6.4
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,2048,0.14841599762439728,0.1407936006784439,0.15617279708385468,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:31,0.6.4
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,3072,0.14550399780273438,0.1382720023393631,0.1539199948310852,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:31,0.6.4
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,4096,0.14467200636863708,0.13808000087738037,0.15168639421463012,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:31,0.6.4
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,5120,0.1457280069589615,0.13897599279880524,0.1539135962724686,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:31,0.6.4
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,6144,0.14667200297117233,0.1394624024629593,0.15324799716472626,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:31,0.6.4
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,7168,0.14979200065135956,0.1430719941854477,0.15615999698638916,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:31,0.6.4
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,8192,0.1499200016260147,0.14345599710941315,0.15614719986915587,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:31,0.6.4
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,9216,0.1518079936504364,0.14513919949531556,0.15914880037307738,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:31,0.6.4
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,10240,0.15292799472808838,0.14597759544849395,0.1614144057035446,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:31,0.6.4
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,11264,0.15059199929237366,0.14905600249767303,0.16228480339050294,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:31,0.6.4
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,12288,0.15756799280643463,0.15663999319076538,0.15862399339675903,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:31,0.6.4
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,13312,0.1871359944343567,0.18614399433135986,0.1881600022315979,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:31,0.6.4
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,14336,0.1924159973859787,0.19145600497722626,0.1934719979763031,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:31,0.6.4
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,15360,0.1974720060825348,0.19644799828529358,0.19862399995326996,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:31,0.6.4
+dyt_beta=False,liger,backward,speed,ms,hidden_size,hidden_size,16384,0.20345599949359894,0.20243200659751892,0.2046079933643341,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:31,0.6.4
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,1024,0.152319997549057,0.15164799988269806,0.15305599570274353,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:39,0.6.4
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,2048,0.2707839906215668,0.27008000016212463,0.2715519964694977,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:39,0.6.4
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,3072,0.39020800590515137,0.38944000005722046,0.3910079896450043,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:39,0.6.4
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,4096,0.49859198927879333,0.4978559911251068,0.4992640018463135,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:39,0.6.4
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,5120,0.6133440136909485,0.6125760078430176,0.6141440272331238,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:39,0.6.4
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,6144,0.716543972492218,0.7158015847206115,0.7174080014228821,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:39,0.6.4
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,7168,0.8275520205497742,0.8267968177795411,0.8283200263977051,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:39,0.6.4
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,8192,0.9326080083847046,0.9318400025367737,0.9335359930992126,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:39,0.6.4
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,9216,1.045151948928833,1.0441279411315918,1.0462975978851319,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:39,0.6.4
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,10240,1.1468479633331299,1.1459200382232666,1.147871971130371,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:39,0.6.4
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,11264,1.2575039863586426,1.2562240362167358,1.25872004032135,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:39,0.6.4
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,12288,1.3627840280532837,1.3616000413894653,1.363916778564453,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:39,0.6.4
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,13312,1.4752320051193237,1.4741439819335938,1.4763200283050537,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:39,0.6.4
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,14336,1.57750403881073,1.5765759944915771,1.5787839889526367,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:39,0.6.4
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,15360,1.6881279945373535,1.6868927717208861,1.689087986946106,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:39,0.6.4
+dyt_beta=False,torch,backward,speed,ms,hidden_size,hidden_size,16384,1.7928159832954407,1.7916287899017336,1.7938239574432373,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:39,0.6.4
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,1024,0.165072001516819,0.15699840188026426,0.18236159682273867,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:53,0.6.4
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,2048,0.169855996966362,0.15922560393810273,0.1930624008178711,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:53,0.6.4
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,3072,0.1627199947834015,0.15513600409030914,0.1741312026977539,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:53,0.6.4
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,4096,0.1613440066576004,0.15441920161247252,0.17027199268341064,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:53,0.6.4
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,5120,0.20099200308322906,0.20025600492954254,0.20182399451732635,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:53,0.6.4
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,6144,0.2319680005311966,0.23097600042819977,0.2327679991722107,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:53,0.6.4
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,7168,0.26281601190567017,0.2620159983634949,0.2635648012161255,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:53,0.6.4
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,8192,0.2927359938621521,0.29203200340270996,0.29340800642967224,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:53,0.6.4
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,9216,1.0456000566482544,1.044550395011902,1.0466879606246948,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:53,0.6.4
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,10240,1.1474239826202393,1.1465280055999756,1.148319959640503,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:53,0.6.4
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,11264,1.257856011390686,1.2564671754837038,1.258944034576416,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:53,0.6.4
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,12288,1.3639999628067017,1.3626879453659058,1.365280032157898,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:53,0.6.4
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,13312,1.474943995475769,1.4739199876785278,1.4759999513626099,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:53,0.6.4
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,14336,1.5774719715118408,1.5763520002365112,1.5785280466079712,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:53,0.6.4
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,15360,1.6879680156707764,1.686835217475891,1.6891200542449951,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:53,0.6.4
+dyt_beta=False,torch_compile,backward,speed,ms,hidden_size,hidden_size,16384,1.7926080226898193,1.7914880514144897,1.7934720516204834,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:28:53,0.6.4
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,1024,0.2537919878959656,0.24294400215148926,0.2673279941082001,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:00,0.6.4
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,2048,0.2503040134906769,0.24076800048351288,0.2640640139579773,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:00,0.6.4
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,3072,0.2524160146713257,0.24329599738121033,0.26556798815727234,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:00,0.6.4
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,4096,0.2555519938468933,0.24540799856185913,0.269651198387146,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:00,0.6.4
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,5120,0.2513599991798401,0.24201600253582,0.2637120008468628,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:00,0.6.4
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,6144,0.25542399287223816,0.2447039932012558,0.2691200077533722,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:00,0.6.4
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,7168,0.24956799298524857,0.2412479966878891,0.25993600487709045,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:00,0.6.4
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,8192,0.24908799678087234,0.24165760576725007,0.25891839861869814,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:00,0.6.4
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,9216,0.2552480101585388,0.24767999351024628,0.26631039977073667,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:00,0.6.4
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,10240,0.25908800959587097,0.2502399981021881,0.27023999094963075,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:00,0.6.4
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,11264,0.2558079957962036,0.2456511974334717,0.26967679262161254,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:00,0.6.4
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,12288,0.25625599920749664,0.2456831932067871,0.27093759179115295,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:00,0.6.4
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,13312,0.2664960026741028,0.2656640112400055,0.2672959864139557,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:00,0.6.4
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,14336,0.27561599016189575,0.27481600642204285,0.27641600370407104,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:00,0.6.4
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,15360,0.287663996219635,0.2867967963218689,0.28846721053123475,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:00,0.6.4
+dyt_beta=False,liger,full,speed,ms,hidden_size,hidden_size,16384,0.29712000489234924,0.29628801345825195,0.29795199632644653,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:00,0.6.4
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,1024,0.19460800290107727,0.19305600225925446,0.1980160027742386,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:09,0.6.4
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,2048,0.36528000235557556,0.3632639944553375,0.368665611743927,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:09,0.6.4
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,3072,0.5249919891357422,0.5229439735412598,0.5278272151947021,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:09,0.6.4
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,4096,0.6760800182819366,0.6739583849906922,0.6788096189498901,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:09,0.6.4
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,5120,0.8264639973640442,0.8243711948394776,0.8289728045463562,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:09,0.6.4
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,6144,0.9748639762401581,0.9725248098373414,0.9775296211242676,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:09,0.6.4
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,7168,1.123408019542694,1.120959997177124,1.126047968864441,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:09,0.6.4
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,8192,1.2715200185775757,1.2690367937088012,1.2738560438156128,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:09,0.6.4
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,9216,1.4193919897079468,1.4171584367752077,1.4221760034561157,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:09,0.6.4
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,10240,1.5666879415512085,1.5644032239913939,1.5692479610443115,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:09,0.6.4
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,11264,1.7149120569229126,1.7127039432525635,1.7176384449005127,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:09,0.6.4
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,12288,1.8633760213851929,1.8610752105712893,1.866207981109619,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:09,0.6.4
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,13312,2.010080099105835,2.0077824115753176,2.012934350967407,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:09,0.6.4
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,14336,2.1577279567718506,2.1557119369506834,2.160633611679077,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:09,0.6.4
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,15360,2.3060801029205322,2.3037951946258546,2.308742332458496,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:09,0.6.4
+dyt_beta=False,torch,full,speed,ms,hidden_size,hidden_size,16384,2.4543039798736572,2.4521920680999756,2.4572992801666262,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:09,0.6.4
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,1024,0.3493119925260544,0.3370176017284393,0.3660863995552063,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:17,0.6.4
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,2048,0.3475840091705322,0.33375999331474304,0.36727039813995366,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:17,0.6.4
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,3072,0.3479520082473755,0.3358847975730896,0.36396161317825315,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:17,0.6.4
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,4096,0.3485279977321625,0.3352319896221161,0.366784006357193,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:17,0.6.4
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,5120,0.3492479920387268,0.3357439935207367,0.36581119894981384,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:17,0.6.4
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,6144,0.35452800989151,0.34296320080757137,0.37255679368972777,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:17,0.6.4
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,7168,0.35577601194381714,0.34219519495964046,0.37092478871345524,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:17,0.6.4
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,8192,0.36428800225257874,0.36233600974082947,0.36579200625419617,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:17,0.6.4
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,9216,1.419711947441101,1.4174079895019531,1.4225280284881592,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:17,0.6.4
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,10240,1.5663360357284546,1.5639551639556886,1.5690367937088012,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:17,0.6.4
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,11264,1.7149120569229126,1.7127615690231321,1.71770236492157,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:17,0.6.4
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,12288,1.8632960319519043,1.860761570930481,1.8661823987960815,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:17,0.6.4
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,13312,2.0100159645080566,2.0079360008239746,2.0129152297973634,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:17,0.6.4
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,14336,2.1576000452041626,2.155340814590454,2.1605952262878416,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:17,0.6.4
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,15360,2.306015968322754,2.3038079738616943,2.3083200454711914,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:17,0.6.4
+dyt_beta=False,torch_compile,full,speed,ms,hidden_size,hidden_size,16384,2.453984022140503,2.451788806915283,2.4569919109344482,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": false}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+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 HBM3,2025-12-19 00:29:17,0.6.4
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,1024,0.013088000006973743,0.012896000407636166,0.013311999849975109,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:28,0.6.4
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,2048,0.018912000581622124,0.018751999363303185,0.019168000668287277,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:28,0.6.4
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,3072,0.024320000782608986,0.024159999564290047,0.02472960017621518,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:28,0.6.4
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,4096,0.031007999554276466,0.030848000198602676,0.03136000037193298,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:28,0.6.4
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,5120,0.03788800165057182,0.037728000432252884,0.03859199956059456,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:28,0.6.4
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,6144,0.04198399931192398,0.041760001331567764,0.042847998440265656,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:28,0.6.4
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,7168,0.04940799996256828,0.0490880012512207,0.05011200159788132,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:28,0.6.4
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,8192,0.05375999957323074,0.053247999399900436,0.05446400120854378,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:28,0.6.4
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,9216,0.060864001512527466,0.060127999633550644,0.06124800071120262,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:28,0.6.4
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,10240,0.0647680014371872,0.06400000303983688,0.06518399715423584,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:28,0.6.4
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,11264,0.07222399860620499,0.0713919997215271,0.07254400104284286,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:28,0.6.4
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,12288,0.0761599987745285,0.07529599964618683,0.07648000121116638,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:28,0.6.4
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,13312,0.08390399813652039,0.08313599973917007,0.08416000008583069,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:28,0.6.4
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,14336,0.08787199854850769,0.08742400258779526,0.0880960002541542,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:28,0.6.4
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,15360,0.09520000219345093,0.09491200000047684,0.09545599669218063,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:28,0.6.4
+dyt_beta=True,liger,forward,speed,ms,hidden_size,hidden_size,16384,0.0992640033364296,0.09894400089979172,0.09955199807882309,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:28,0.6.4
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,1024,0.0615679994225502,0.059776000678539276,0.062111999839544296,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:37,0.6.4
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,2048,0.12700800597667694,0.1252799928188324,0.1276160031557083,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:37,0.6.4
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,3072,0.18297599256038666,0.1812479943037033,0.18358400464057922,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:37,0.6.4
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,4096,0.236735999584198,0.235167995095253,0.23740799725055695,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:37,0.6.4
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,5120,0.2903999984264374,0.28867200016975403,0.29097598791122437,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:37,0.6.4
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,6144,0.3441919982433319,0.34275200963020325,0.34508800506591797,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:37,0.6.4
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,7168,0.39817601442337036,0.3962559998035431,0.3986560106277466,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:37,0.6.4
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,8192,0.45184001326560974,0.4501760005950928,0.45257601141929626,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:37,0.6.4
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,9216,0.5055680274963379,0.5035520195960999,0.5060160160064697,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:37,0.6.4
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,10240,0.5591999888420105,0.5575680136680603,0.559935986995697,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:37,0.6.4
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,11264,0.6128320097923279,0.6110399961471558,0.6133760213851929,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:37,0.6.4
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,12288,0.666592001914978,0.6651584148406982,0.6673920154571533,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:37,0.6.4
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,13312,0.7204160094261169,0.718720018863678,0.7209920287132263,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:37,0.6.4
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,14336,0.7744320034980774,0.772876787185669,0.7752959728240967,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:37,0.6.4
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,15360,0.8281279802322388,0.8264511942863464,0.8287360072135925,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:37,0.6.4
+dyt_beta=True,torch,forward,speed,ms,hidden_size,hidden_size,16384,0.8819519877433777,0.8802239894866943,0.8825600147247314,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:37,0.6.4
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,1024,0.06143999844789505,0.05967999994754791,0.06204799935221672,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:45,0.6.4
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,2048,0.1271360069513321,0.12534399330615997,0.12771199643611908,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:45,0.6.4
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,3072,0.18303999304771423,0.18121600151062012,0.18367999792099,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:45,0.6.4
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,4096,0.23683199286460876,0.23523199558258057,0.2375359982252121,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:45,0.6.4
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,5120,0.29049599170684814,0.2884800136089325,0.29100799560546875,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:45,0.6.4
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,6144,0.34404799342155457,0.34252798557281494,0.3449600040912628,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:45,0.6.4
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,7168,0.398144006729126,0.39612799882888794,0.3986560106277466,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:45,0.6.4
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,8192,0.45184001326560974,0.45020800828933716,0.4525440037250519,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:45,0.6.4
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,9216,0.5054720044136047,0.5036479830741882,0.5059840083122253,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:45,0.6.4
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,10240,0.5591519773006439,0.5575616121292114,0.5600000023841858,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:45,0.6.4
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,11264,0.6128640174865723,0.6109439730644226,0.6134080290794373,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:45,0.6.4
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,12288,0.6665599942207336,0.6649280190467834,0.6674559712409973,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:45,0.6.4
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,13312,0.7204160094261169,0.7185919880867004,0.7209600210189819,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:45,0.6.4
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,14336,0.774399995803833,0.7728000283241272,0.7752640247344971,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:45,0.6.4
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,15360,0.8280640244483948,0.826367974281311,0.8286719918251038,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:45,0.6.4
+dyt_beta=True,torch_compile,forward,speed,ms,hidden_size,hidden_size,16384,0.8819839954376221,0.8803200125694275,0.882643175125122,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:45,0.6.4
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,1024,0.19307199865579605,0.18566399812698364,0.20163199305534363,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:55,0.6.4
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,2048,0.19225600361824036,0.18489600718021393,0.20003199577331543,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:55,0.6.4
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,3072,0.19203199446201324,0.18456960320472715,0.2016959935426712,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:55,0.6.4
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,4096,0.19142399728298187,0.18457600474357605,0.19840000569820404,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:55,0.6.4
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,5120,0.1939840018749237,0.1870719939470291,0.20121599733829498,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:55,0.6.4
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,6144,0.19180800020694733,0.1844799965620041,0.2006207972764969,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:55,0.6.4
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,7168,0.19312000274658203,0.1861567974090576,0.20179200172424316,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:55,0.6.4
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,8192,0.19352000206708908,0.1872255951166153,0.2006399929523468,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:55,0.6.4
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,9216,0.19699199497699738,0.19090559482574462,0.20710399746894836,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:55,0.6.4
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,10240,0.19728000462055206,0.19069440066814422,0.20386559665203094,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:55,0.6.4
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,11264,0.19788800179958344,0.1910975992679596,0.20568959414958954,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:55,0.6.4
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,12288,0.19644799828529358,0.18771199882030487,0.20531199872493744,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:55,0.6.4
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,13312,0.2083360031247139,0.20735999941825867,0.20924800634384155,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:55,0.6.4
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,14336,0.21350400149822235,0.21264000236988068,0.21436800062656403,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:55,0.6.4
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,15360,0.2189120054244995,0.21795199811458588,0.21990400552749634,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:55,0.6.4
+dyt_beta=True,liger,backward,speed,ms,hidden_size,hidden_size,16384,0.2242880016565323,0.22339199483394623,0.22518399357795715,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:29:55,0.6.4
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,1024,0.16512000560760498,0.16435199975967407,0.1658879965543747,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:03,0.6.4
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,2048,0.29552000761032104,0.29475200176239014,0.2961919903755188,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:03,0.6.4
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,3072,0.4172160029411316,0.4163520038127899,0.4182080030441284,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:03,0.6.4
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,4096,0.5407360196113586,0.5398143768310547,0.5415679812431335,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:03,0.6.4
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,5120,0.6527040004730225,0.65174400806427,0.6536639928817749,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:03,0.6.4
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,6144,0.7712960243225098,0.7703040242195129,0.7723519802093506,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:03,0.6.4
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,7168,0.881600022315979,0.8806271910667419,0.8825600147247314,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:03,0.6.4
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,8192,1.0000959634780884,0.9991359710693359,1.0009855508804322,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:03,0.6.4
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,9216,1.1073280572891235,1.1062400341033936,1.1083199977874756,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:03,0.6.4
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,10240,1.225343942642212,1.224128007888794,1.2264000177383423,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:03,0.6.4
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,11264,1.3328160047531128,1.3317760229110718,1.3338559865951538,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:03,0.6.4
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,12288,1.4494719505310059,1.4481600522994995,1.450713562965393,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:03,0.6.4
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,13312,1.5577280521392822,1.5567679405212402,1.558784008026123,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:03,0.6.4
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,14336,1.6752959489822388,1.6741119623184204,1.6764479875564575,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:03,0.6.4
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,15360,1.7834880352020264,1.7819520235061646,1.7845439910888672,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:03,0.6.4
+dyt_beta=True,torch,backward,speed,ms,hidden_size,hidden_size,16384,1.8999040126800537,1.8987583875656127,1.9011456012725831,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:03,0.6.4
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,1024,0.1650560051202774,0.16419200599193573,0.16579200327396393,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:11,0.6.4
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,2048,0.2954240143299103,0.29465600848197937,0.2961919903755188,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:11,0.6.4
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,3072,0.4169920086860657,0.4161919951438904,0.4180159866809845,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:11,0.6.4
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,4096,0.5406720042228699,0.5397120118141174,0.5415935873985291,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:11,0.6.4
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,5120,0.6526399850845337,0.65174400806427,0.6535999774932861,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:11,0.6.4
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,6144,0.7711359858512878,0.7700863718986511,0.7720959782600403,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:11,0.6.4
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,7168,0.8815680146217346,0.880511999130249,0.8825600147247314,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:11,0.6.4
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,8192,1.0000319480895996,0.9989759922027588,1.0008959770202637,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:11,0.6.4
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,9216,1.1073600053787231,1.1062079668045044,1.1084480285644531,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:11,0.6.4
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,10240,1.2254079580307007,1.2241920232772827,1.2264000177383423,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:11,0.6.4
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,11264,1.332576036453247,1.331558418273926,1.3336960077285767,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:11,0.6.4
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,12288,1.4495359659194946,1.4484543561935426,1.450809621810913,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:11,0.6.4
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,13312,1.5577600002288818,1.5565760135650635,1.5587199926376343,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:11,0.6.4
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,14336,1.675104022026062,1.6737983942031862,1.6763839721679688,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:11,0.6.4
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,15360,1.7832000255584717,1.7818880081176758,1.7845439910888672,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:11,0.6.4
+dyt_beta=True,torch_compile,backward,speed,ms,hidden_size,hidden_size,16384,1.8998719453811646,1.898848056793213,1.9010560512542725,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:11,0.6.4
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,1024,0.2895359992980957,0.27990400791168213,0.29949440360069274,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:19,0.6.4
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,2048,0.29075199365615845,0.2812480032444,0.30244480371475224,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:19,0.6.4
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,3072,0.2902719974517822,0.2810815930366516,0.3011583924293518,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:19,0.6.4
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,4096,0.29151999950408936,0.2810559868812561,0.3047040104866028,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:19,0.6.4
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,5120,0.2911359965801239,0.28332799673080444,0.30167039632797243,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:19,0.6.4
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,6144,0.2935999929904938,0.2829952001571655,0.3110208094120026,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:19,0.6.4
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,7168,0.29049599170684814,0.28114560842514036,0.30209920406341556,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:19,0.6.4
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,8192,0.28803199529647827,0.2784000039100647,0.3001280128955841,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:19,0.6.4
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,9216,0.29260799288749695,0.2826176047325134,0.30688639283180236,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:19,0.6.4
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,10240,0.29155200719833374,0.2821120023727417,0.3031359910964966,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:19,0.6.4
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,11264,0.291920006275177,0.2847679853439331,0.3019327998161316,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:19,0.6.4
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,12288,0.29395200312137604,0.285504013299942,0.3073599934577942,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:19,0.6.4
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,13312,0.29497599601745605,0.2866879999637604,0.30967039465904234,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:19,0.6.4
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,14336,0.29766398668289185,0.29526400566101074,0.29926401376724243,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:19,0.6.4
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,15360,0.3110080063343048,0.31007999181747437,0.3118720054626465,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:19,0.6.4
+dyt_beta=True,liger,full,speed,ms,hidden_size,hidden_size,16384,0.3210560083389282,0.3203519880771637,0.32182401418685913,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:19,0.6.4
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,1024,0.22336000204086304,0.2224320024251938,0.2242240011692047,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:27,0.6.4
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,2048,0.4192639887332916,0.41839998960494995,0.4201599955558777,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:27,0.6.4
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,3072,0.5976639986038208,0.5966463923454285,0.5986559987068176,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:27,0.6.4
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,4096,0.7745599746704102,0.7735360264778137,0.775551974773407,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:27,0.6.4
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,5120,0.9399679899215698,0.9390592098236084,0.9410560131072998,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:27,0.6.4
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,6144,1.112671971321106,1.1116991996765138,1.1135679483413696,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:27,0.6.4
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,7168,1.2762240171432495,1.27510404586792,1.2773760557174683,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:27,0.6.4
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,8192,1.449120044708252,1.4481279850006104,1.4501440525054932,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:27,0.6.4
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,9216,1.6099200248718262,1.6087039709091187,1.6110591650009156,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:27,0.6.4
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,10240,1.781823992729187,1.7807168245315552,1.7830079793930054,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:27,0.6.4
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,11264,1.9425280094146729,1.9413119554519653,1.9438079595565796,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:27,0.6.4
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,12288,2.1135679483413696,2.1123327255249023,2.1146559715270996,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:27,0.6.4
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,13312,2.275376081466675,2.2740287303924562,2.2769407272338866,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:27,0.6.4
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,14336,2.4469759464263916,2.4454911708831784,2.448428773880005,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:27,0.6.4
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,15360,2.607935905456543,2.6067711353302,2.6090240478515625,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:27,0.6.4
+dyt_beta=True,torch,full,speed,ms,hidden_size,hidden_size,16384,2.7798080444335938,2.7783104896545407,2.7810495376586917,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:27,0.6.4
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,1024,0.22300800681114197,0.22118400037288666,0.23195519745349885,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:36,0.6.4
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,2048,0.419295996427536,0.4184640049934387,0.42025598883628845,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:36,0.6.4
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,3072,0.5973439812660217,0.5963775873184204,0.5983999967575073,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:36,0.6.4
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,4096,0.7746559977531433,0.7735679745674133,0.7757440209388733,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:36,0.6.4
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,5120,0.9399679899215698,0.9390079975128174,0.9409599900245667,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:36,0.6.4
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,6144,1.1127680540084839,1.1117759943008423,1.1136512041091917,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:36,0.6.4
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,7168,1.2766079902648926,1.2755327939987182,1.2776639461517334,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:36,0.6.4
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,8192,1.4492160081863403,1.4481600522994995,1.4501503944396972,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:36,0.6.4
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,9216,1.6095679998397827,1.608396792411804,1.6106751918792725,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:36,0.6.4
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,10240,1.7813119888305664,1.7802239656448364,1.7823679447174072,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:36,0.6.4
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,11264,1.9425119757652283,1.941472053527832,1.9438079595565796,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:36,0.6.4
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,12288,2.1132800579071045,2.1120895385742187,2.114579200744629,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:36,0.6.4
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,13312,2.275023937225342,2.2737600803375244,2.276089668273926,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:36,0.6.4
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,14336,2.4463679790496826,2.4449151515960694,2.4476927757263183,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:36,0.6.4
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,15360,2.6076159477233887,2.6065663814544675,2.608704090118408,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:36,0.6.4
+dyt_beta=True,torch_compile,full,speed,ms,hidden_size,hidden_size,16384,2.7799038887023926,2.7782591819763183,2.781183958053589,"{""BT"": 4096, ""dtype"": ""torch.bfloat16"", ""beta"": true}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+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 HBM3,2025-12-19 00:30:36,0.6.4
+embedding,liger,forward,speed,ms,V,embedding dimension,1024,0.024639999493956566,0.024320000782608986,0.025062399357557295,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:48,0.6.4
+embedding,liger,forward,speed,ms,V,embedding dimension,2048,0.026079999282956123,0.025728000327944756,0.026559999212622643,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:48,0.6.4
+embedding,liger,forward,speed,ms,V,embedding dimension,4096,0.028031999245285988,0.02778879925608635,0.028518399968743325,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:48,0.6.4
+embedding,liger,forward,speed,ms,V,embedding dimension,8192,0.031136000528931618,0.0307839997112751,0.031808000057935715,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:48,0.6.4
+embedding,liger,forward,speed,ms,V,embedding dimension,16384,0.03510399907827377,0.03471999987959862,0.035840000957250595,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:48,0.6.4
+embedding,liger,forward,speed,ms,V,embedding dimension,32768,0.03782400116324425,0.03734400123357773,0.038495998829603195,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:48,0.6.4
+embedding,liger,forward,speed,ms,V,embedding dimension,65536,0.039744000881910324,0.03923200070858002,0.04037119969725608,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:48,0.6.4
+embedding,liger,forward,speed,ms,V,embedding dimension,131072,0.04118400067090988,0.040608000010252,0.04182400181889534,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:48,0.6.4
+embedding,huggingface,forward,speed,ms,V,embedding dimension,1024,0.02457600086927414,0.024351999163627625,0.02505600079894066,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:51,0.6.4
+embedding,huggingface,forward,speed,ms,V,embedding dimension,2048,0.02691200003027916,0.026623999699950218,0.027488000690937042,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:51,0.6.4
+embedding,huggingface,forward,speed,ms,V,embedding dimension,4096,0.03094400092959404,0.0306560005992651,0.0315839983522892,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:51,0.6.4
+embedding,huggingface,forward,speed,ms,V,embedding dimension,8192,0.03558399900794029,0.035232000052928925,0.03625600039958954,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:51,0.6.4
+embedding,huggingface,forward,speed,ms,V,embedding dimension,16384,0.038656000047922134,0.03830400109291077,0.039455998688936234,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:51,0.6.4
+embedding,huggingface,forward,speed,ms,V,embedding dimension,32768,0.04041599854826927,0.040031999349594116,0.041152000427246094,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:51,0.6.4
+embedding,huggingface,forward,speed,ms,V,embedding dimension,65536,0.041471999138593674,0.041075199842453,0.04227200150489807,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:51,0.6.4
+embedding,huggingface,forward,speed,ms,V,embedding dimension,131072,0.04224000126123428,0.041919998824596405,0.04307200014591217,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:30:51,0.6.4
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,1024,0.027807999402284622,0.02703999914228916,0.029152000322937965,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:00,0.6.4
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,2048,0.028031999245285988,0.027187200263142586,0.02919680066406727,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:00,0.6.4
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,4096,0.03017600066959858,0.029952000826597214,0.031065599992871284,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:00,0.6.4
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,8192,0.03468799963593483,0.03440000116825104,0.03548799827694893,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:00,0.6.4
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,16384,0.03791999816894531,0.037567999213933945,0.038656000047922134,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:00,0.6.4
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,32768,0.039744000881910324,0.03936000168323517,0.04047999903559685,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:00,0.6.4
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,65536,0.04092799872159958,0.04047999903559685,0.0416640006005764,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:00,0.6.4
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,131072,0.041519999504089355,0.041120000183582306,0.04227200150489807,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:00,0.6.4
+embedding,liger,backward,speed,ms,V,embedding dimension,1024,0.14137599617242813,0.14047999680042267,0.14236800372600555,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:03,0.6.4
+embedding,liger,backward,speed,ms,V,embedding dimension,2048,0.14735999703407288,0.14604159593582153,0.1480959951877594,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:03,0.6.4
+embedding,liger,backward,speed,ms,V,embedding dimension,4096,0.15779200196266174,0.15615999698638916,0.15852800011634827,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:03,0.6.4
+embedding,liger,backward,speed,ms,V,embedding dimension,8192,0.18171200156211853,0.18028160035610197,0.182559996843338,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:03,0.6.4
+embedding,liger,backward,speed,ms,V,embedding dimension,16384,0.2237440049648285,0.22230400145053864,0.22467200458049774,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:03,0.6.4
+embedding,liger,backward,speed,ms,V,embedding dimension,32768,0.2943359911441803,0.2928831994533539,0.2952127933502197,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:03,0.6.4
+embedding,liger,backward,speed,ms,V,embedding dimension,65536,0.42630401253700256,0.42481918931007384,0.42720639109611513,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:03,0.6.4
+embedding,liger,backward,speed,ms,V,embedding dimension,131072,0.687248021364212,0.6859968185424805,0.688319981098175,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:03,0.6.4
+embedding,huggingface,backward,speed,ms,V,embedding dimension,1024,0.20374399423599243,0.20307199656963348,0.20455679595470427,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:06,0.6.4
+embedding,huggingface,backward,speed,ms,V,embedding dimension,2048,0.21513600647449493,0.2144639939069748,0.2160000056028366,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:06,0.6.4
+embedding,huggingface,backward,speed,ms,V,embedding dimension,4096,0.24646399915218353,0.245728000998497,0.2473600059747696,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:06,0.6.4
+embedding,huggingface,backward,speed,ms,V,embedding dimension,8192,0.31436800956726074,0.31382399797439575,0.3152639865875244,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:06,0.6.4
+embedding,huggingface,backward,speed,ms,V,embedding dimension,16384,0.4148000031709671,0.41391358971595765,0.4157888114452362,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:06,0.6.4
+embedding,huggingface,backward,speed,ms,V,embedding dimension,32768,0.5061439871788025,0.5053120255470276,0.5070847749710082,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:06,0.6.4
+embedding,huggingface,backward,speed,ms,V,embedding dimension,65536,0.6488160192966461,0.6478400230407715,0.6499840021133423,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:06,0.6.4
+embedding,huggingface,backward,speed,ms,V,embedding dimension,131072,0.9160000085830688,0.9150272130966187,0.917030394077301,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:06,0.6.4
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,1024,0.11604799702763557,0.10487040132284164,0.12928640246391296,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:13,0.6.4
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,2048,0.10875200107693672,0.10172799974679947,0.11568000167608261,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:13,0.6.4
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,4096,0.11468800157308578,0.10778240114450455,0.12522239983081818,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:13,0.6.4
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,8192,0.11660800129175186,0.10922879725694656,0.12883199751377106,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:13,0.6.4
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,16384,0.15455999970436096,0.15322880446910858,0.15545600652694702,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:13,0.6.4
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,32768,0.2271680012345314,0.22550399601459503,0.2279359996318817,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:13,0.6.4
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,65536,0.3611840009689331,0.35959680676460265,0.3622080087661743,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:13,0.6.4
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,131072,0.6215360164642334,0.6198400259017944,0.6226943969726563,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:13,0.6.4
+embedding,liger,full,speed,ms,V,embedding dimension,1024,0.16524800658226013,0.16467200219631195,0.16582399606704712,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:16,0.6.4
+embedding,liger,full,speed,ms,V,embedding dimension,2048,0.17017599940299988,0.1690816044807434,0.171424001455307,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:16,0.6.4
+embedding,liger,full,speed,ms,V,embedding dimension,4096,0.1839200034737587,0.18335359692573547,0.18457600474357605,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:16,0.6.4
+embedding,liger,full,speed,ms,V,embedding dimension,8192,0.21119999885559082,0.21062399446964264,0.2120320051908493,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:16,0.6.4
+embedding,liger,full,speed,ms,V,embedding dimension,16384,0.25702399015426636,0.2563199996948242,0.2577983975410461,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:16,0.6.4
+embedding,liger,full,speed,ms,V,embedding dimension,32768,0.32902398705482483,0.3279680013656616,0.3302080035209656,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:16,0.6.4
+embedding,liger,full,speed,ms,V,embedding dimension,65536,0.4646719992160797,0.46323201060295105,0.4658431947231293,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:16,0.6.4
+embedding,liger,full,speed,ms,V,embedding dimension,131072,0.7251200079917908,0.724128007888794,0.726265585422516,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:16,0.6.4
+embedding,huggingface,full,speed,ms,V,embedding dimension,1024,0.2263839989900589,0.22553600370883942,0.22732800245285034,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:19,0.6.4
+embedding,huggingface,full,speed,ms,V,embedding dimension,2048,0.23958399891853333,0.2388480007648468,0.24051199853420258,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:19,0.6.4
+embedding,huggingface,full,speed,ms,V,embedding dimension,4096,0.2746880054473877,0.2739199995994568,0.27555200457572937,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:19,0.6.4
+embedding,huggingface,full,speed,ms,V,embedding dimension,8192,0.34652799367904663,0.34544000029563904,0.34751999378204346,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:19,0.6.4
+embedding,huggingface,full,speed,ms,V,embedding dimension,16384,0.4530719965696335,0.4521600008010864,0.4541440010070801,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:19,0.6.4
+embedding,huggingface,full,speed,ms,V,embedding dimension,32768,0.5441920161247253,0.5434752225875854,0.5450687885284424,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:19,0.6.4
+embedding,huggingface,full,speed,ms,V,embedding dimension,65536,0.6883839964866638,0.6874752163887023,0.689036774635315,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:19,0.6.4
+embedding,huggingface,full,speed,ms,V,embedding dimension,131072,0.957616001367569,0.9567040205001831,0.9587200284004211,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:19,0.6.4
+embedding,torch_compile,full,speed,ms,V,embedding dimension,1024,0.2667199969291687,0.25571200251579285,0.2802880108356476,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:22,0.6.4
+embedding,torch_compile,full,speed,ms,V,embedding dimension,2048,0.2613440006971359,0.25225600600242615,0.2727360129356384,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:22,0.6.4
+embedding,torch_compile,full,speed,ms,V,embedding dimension,4096,0.26601600646972656,0.2532544076442719,0.27712640166282654,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:22,0.6.4
+embedding,torch_compile,full,speed,ms,V,embedding dimension,8192,0.2640320062637329,0.2546048045158386,0.27811199426651,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:22,0.6.4
+embedding,torch_compile,full,speed,ms,V,embedding dimension,16384,0.2595200091600418,0.25068798661231995,0.26979199051856995,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:22,0.6.4
+embedding,torch_compile,full,speed,ms,V,embedding dimension,32768,0.2736479938030243,0.26347520351409914,0.2932287991046906,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:22,0.6.4
+embedding,torch_compile,full,speed,ms,V,embedding dimension,65536,0.3999679982662201,0.39923200011253357,0.400736004114151,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:22,0.6.4
+embedding,torch_compile,full,speed,ms,V,embedding dimension,131072,0.6621119976043701,0.6611840128898621,0.6631360054016113,"{""B"": 32, ""T"": 512, ""D"": 768, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:22,0.6.4
+embedding,liger,forward,speed,ms,V,embedding dimension,1024,0.10630399733781815,0.1058880016207695,0.10681600123643875,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:33,0.6.4
+embedding,liger,forward,speed,ms,V,embedding dimension,2048,0.11404799669981003,0.11366400122642517,0.11446399986743927,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:33,0.6.4
+embedding,liger,forward,speed,ms,V,embedding dimension,4096,0.1260800063610077,0.12569600343704224,0.12665599584579468,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:33,0.6.4
+embedding,liger,forward,speed,ms,V,embedding dimension,8192,0.14396800100803375,0.14348800480365753,0.14470399916172028,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:33,0.6.4
+embedding,liger,forward,speed,ms,V,embedding dimension,16384,0.16371199488639832,0.16291199624538422,0.16432000696659088,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:33,0.6.4
+embedding,liger,forward,speed,ms,V,embedding dimension,32768,0.177279993891716,0.17660799622535706,0.17783040404319764,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:33,0.6.4
+embedding,liger,forward,speed,ms,V,embedding dimension,65536,0.1870719939470291,0.18652799725532532,0.18751999735832214,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:33,0.6.4
+embedding,liger,forward,speed,ms,V,embedding dimension,131072,0.19196799397468567,0.19148799777030945,0.19232640266418458,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:33,0.6.4
+embedding,huggingface,forward,speed,ms,V,embedding dimension,1024,0.10838399827480316,0.10809600353240967,0.10876800119876862,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:45,0.6.4
+embedding,huggingface,forward,speed,ms,V,embedding dimension,2048,0.12457600235939026,0.12422399967908859,0.12507520318031312,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:45,0.6.4
+embedding,huggingface,forward,speed,ms,V,embedding dimension,4096,0.14713600277900696,0.1467711955308914,0.14787200093269348,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:45,0.6.4
+embedding,huggingface,forward,speed,ms,V,embedding dimension,8192,0.1679680049419403,0.16726399958133698,0.16828800737857819,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:45,0.6.4
+embedding,huggingface,forward,speed,ms,V,embedding dimension,16384,0.18057599663734436,0.179967999458313,0.18092800676822662,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:45,0.6.4
+embedding,huggingface,forward,speed,ms,V,embedding dimension,32768,0.18688000738620758,0.18639999628067017,0.18726399540901184,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:45,0.6.4
+embedding,huggingface,forward,speed,ms,V,embedding dimension,65536,0.19148799777030945,0.191103994846344,0.19183999300003052,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:45,0.6.4
+embedding,huggingface,forward,speed,ms,V,embedding dimension,131072,0.1932159960269928,0.19289599359035492,0.1934719979763031,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:45,0.6.4
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,1024,0.10831999778747559,0.10796800255775452,0.10870400071144104,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:56,0.6.4
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,2048,0.12489599734544754,0.12460800260305405,0.12546559572219848,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:56,0.6.4
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,4096,0.14723199605941772,0.14681600034236908,0.14795519709587096,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:56,0.6.4
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,8192,0.16790400445461273,0.1671999990940094,0.16838400065898895,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:56,0.6.4
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,16384,0.18140800297260284,0.18086400628089905,0.18169599771499634,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:56,0.6.4
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,32768,0.18742400407791138,0.18694399297237396,0.18777599930763245,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:56,0.6.4
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,65536,0.19206400215625763,0.19171200692653656,0.19235199689865112,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:56,0.6.4
+embedding,torch_compile,forward,speed,ms,V,embedding dimension,131072,0.19388799369335175,0.19353599846363068,0.194240003824234,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:31:56,0.6.4
+embedding,liger,backward,speed,ms,V,embedding dimension,1024,0.6785600185394287,0.6772223949432373,0.6793599724769592,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:07,0.6.4
+embedding,liger,backward,speed,ms,V,embedding dimension,2048,0.714464008808136,0.713043224811554,0.7152704119682312,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:07,0.6.4
+embedding,liger,backward,speed,ms,V,embedding dimension,4096,0.7830079793930054,0.7817216277122497,0.7841280102729797,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:07,0.6.4
+embedding,liger,backward,speed,ms,V,embedding dimension,8192,0.9008319973945618,0.8992639780044556,0.9017919898033142,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:07,0.6.4
+embedding,liger,backward,speed,ms,V,embedding dimension,16384,1.1001280546188354,1.098803210258484,1.1012608051300048,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:07,0.6.4
+embedding,liger,backward,speed,ms,V,embedding dimension,32768,1.46561598777771,1.464147186279297,1.466758370399475,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:07,0.6.4
+embedding,liger,backward,speed,ms,V,embedding dimension,65536,2.165839910507202,2.164672040939331,2.1669952392578127,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:07,0.6.4
+embedding,liger,backward,speed,ms,V,embedding dimension,131072,3.548192024230957,3.5469439029693604,3.550528049468994,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:07,0.6.4
+embedding,huggingface,backward,speed,ms,V,embedding dimension,1024,0.4800319969654083,0.4792703926563263,0.48076799511909485,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:19,0.6.4
+embedding,huggingface,backward,speed,ms,V,embedding dimension,2048,0.5436480045318604,0.5428095817565918,0.5445503830909729,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:19,0.6.4
+embedding,huggingface,backward,speed,ms,V,embedding dimension,4096,0.692799985408783,0.6916928172111511,0.6936960220336914,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:19,0.6.4
+embedding,huggingface,backward,speed,ms,V,embedding dimension,8192,0.9764159917831421,0.9751423835754394,0.9777920246124268,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:19,0.6.4
+embedding,huggingface,backward,speed,ms,V,embedding dimension,16384,1.424064040184021,1.4229439496994019,1.4251519441604614,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:19,0.6.4
+embedding,huggingface,backward,speed,ms,V,embedding dimension,32768,1.8586560487747192,1.857408046722412,1.8598719835281372,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:19,0.6.4
+embedding,huggingface,backward,speed,ms,V,embedding dimension,65536,2.597440004348755,2.5961663246154782,2.59935359954834,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:19,0.6.4
+embedding,huggingface,backward,speed,ms,V,embedding dimension,131072,4.009552001953125,4.007353687286377,4.0120831489562985,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:19,0.6.4
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,1024,0.47944000363349915,0.47859200835227966,0.4803839921951294,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:30,0.6.4
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,2048,0.543615996837616,0.542579185962677,0.5446079969406128,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:30,0.6.4
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,4096,0.6911680102348328,0.6903807997703553,0.6923840045928955,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:30,0.6.4
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,8192,0.9764800071716309,0.9753343939781188,0.977728009223938,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:30,0.6.4
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,16384,1.4209280014038086,1.4196799993515015,1.4222079515457153,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:30,0.6.4
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,32768,1.8570560216903687,1.8554879426956177,1.8581759929656982,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:30,0.6.4
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,65536,2.601439952850342,2.5998208045959474,2.6027904033660887,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:30,0.6.4
+embedding,torch_compile,backward,speed,ms,V,embedding dimension,131072,4.0110719203948975,4.009414386749267,4.0136768341064455,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:30,0.6.4
+embedding,liger,full,speed,ms,V,embedding dimension,1024,0.7821440100669861,0.7810879945755005,0.7829823732376099,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:41,0.6.4
+embedding,liger,full,speed,ms,V,embedding dimension,2048,0.8256000280380249,0.8244671940803527,0.8265088081359864,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:41,0.6.4
+embedding,liger,full,speed,ms,V,embedding dimension,4096,0.9063680171966553,0.9051839709281921,0.9075199961662292,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:41,0.6.4
+embedding,liger,full,speed,ms,V,embedding dimension,8192,1.042688012123108,1.041484785079956,1.04443519115448,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:41,0.6.4
+embedding,liger,full,speed,ms,V,embedding dimension,16384,1.2611039876937866,1.2595392227172852,1.2620927810668945,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:41,0.6.4
+embedding,liger,full,speed,ms,V,embedding dimension,32768,1.639311969280243,1.6382399797439575,1.6400320529937744,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:41,0.6.4
+embedding,liger,full,speed,ms,V,embedding dimension,65536,2.348191976547241,2.3469440937042236,2.3493120670318604,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:41,0.6.4
+embedding,liger,full,speed,ms,V,embedding dimension,131072,3.7395360469818115,3.737823963165283,3.7418880462646484,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:41,0.6.4
+embedding,huggingface,full,speed,ms,V,embedding dimension,1024,0.5860480070114136,0.585536003112793,0.5867968082427979,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:52,0.6.4
+embedding,huggingface,full,speed,ms,V,embedding dimension,2048,0.6661760210990906,0.6654208183288575,0.6669440269470215,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:52,0.6.4
+embedding,huggingface,full,speed,ms,V,embedding dimension,4096,0.8377759754657745,0.837004804611206,0.8386752128601074,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:52,0.6.4
+embedding,huggingface,full,speed,ms,V,embedding dimension,8192,1.1393280029296875,1.138335943222046,1.140671968460083,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:52,0.6.4
+embedding,huggingface,full,speed,ms,V,embedding dimension,16384,1.5993280410766602,1.597708797454834,1.6007039546966553,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:52,0.6.4
+embedding,huggingface,full,speed,ms,V,embedding dimension,32768,2.0393919944763184,2.038259172439575,2.0409087181091308,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:52,0.6.4
+embedding,huggingface,full,speed,ms,V,embedding dimension,65536,2.7899359464645386,2.7884544372558597,2.791865587234497,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:52,0.6.4
+embedding,huggingface,full,speed,ms,V,embedding dimension,131072,4.200319766998291,4.198073673248291,4.201926422119141,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:32:52,0.6.4
+embedding,torch_compile,full,speed,ms,V,embedding dimension,1024,0.5855199992656708,0.5848448157310486,0.5861759781837463,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:04,0.6.4
+embedding,torch_compile,full,speed,ms,V,embedding dimension,2048,0.6654559969902039,0.6649280190467834,0.6662335872650146,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:04,0.6.4
+embedding,torch_compile,full,speed,ms,V,embedding dimension,4096,0.8355519771575928,0.8344256043434144,0.8367807984352111,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:04,0.6.4
+embedding,torch_compile,full,speed,ms,V,embedding dimension,8192,1.14409601688385,1.1431360244750977,1.145535945892334,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:04,0.6.4
+embedding,torch_compile,full,speed,ms,V,embedding dimension,16384,1.5953919887542725,1.5937663793563843,1.5972607851028442,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:04,0.6.4
+embedding,torch_compile,full,speed,ms,V,embedding dimension,32768,2.0385921001434326,2.0372543334960938,2.0400959968566896,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:04,0.6.4
+embedding,torch_compile,full,speed,ms,V,embedding dimension,65536,2.7920639514923096,2.790463924407959,2.7938239574432373,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:04,0.6.4
+embedding,torch_compile,full,speed,ms,V,embedding dimension,131072,4.201375961303711,4.200735950469971,4.202848052978515,"{""B"": 8, ""T"": 2048, ""D"": 4096, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:04,0.6.4
+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 HBM3,2025-12-19 00:33:06,0.6.4
+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 HBM3,2025-12-19 00:33:06,0.6.4
+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 HBM3,2025-12-19 00:33:06,0.6.4
+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 HBM3,2025-12-19 00:33:06,0.6.4
+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 HBM3,2025-12-19 00:33:06,0.6.4
+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 HBM3,2025-12-19 00:33:06,0.6.4
+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 HBM3,2025-12-19 00:33:06,0.6.4
+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 HBM3,2025-12-19 00:33:06,0.6.4
+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 HBM3,2025-12-19 00:33:08,0.6.4
+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 HBM3,2025-12-19 00:33:08,0.6.4
+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 HBM3,2025-12-19 00:33:08,0.6.4
+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 HBM3,2025-12-19 00:33:08,0.6.4
+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 HBM3,2025-12-19 00:33:08,0.6.4
+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 HBM3,2025-12-19 00:33:08,0.6.4
+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 HBM3,2025-12-19 00:33:08,0.6.4
+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 HBM3,2025-12-19 00:33:08,0.6.4
+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 HBM3,2025-12-19 00:33:10,0.6.4
+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 HBM3,2025-12-19 00:33:10,0.6.4
+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 HBM3,2025-12-19 00:33:10,0.6.4
+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 HBM3,2025-12-19 00:33:10,0.6.4
+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 HBM3,2025-12-19 00:33:10,0.6.4
+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 HBM3,2025-12-19 00:33:10,0.6.4
+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 HBM3,2025-12-19 00:33:10,0.6.4
+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 HBM3,2025-12-19 00:33:10,0.6.4
+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 HBM3,2025-12-19 00:33:20,0.6.4
+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 HBM3,2025-12-19 00:33:20,0.6.4
+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 HBM3,2025-12-19 00:33:20,0.6.4
+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 HBM3,2025-12-19 00:33:20,0.6.4
+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 HBM3,2025-12-19 00:33:20,0.6.4
+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 HBM3,2025-12-19 00:33:20,0.6.4
+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 HBM3,2025-12-19 00:33:20,0.6.4
+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 HBM3,2025-12-19 00:33:20,0.6.4
+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 HBM3,2025-12-19 00:33:31,0.6.4
+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 HBM3,2025-12-19 00:33:31,0.6.4
+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 HBM3,2025-12-19 00:33:31,0.6.4
+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 HBM3,2025-12-19 00:33:31,0.6.4
+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 HBM3,2025-12-19 00:33:31,0.6.4
+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 HBM3,2025-12-19 00:33:31,0.6.4
+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 HBM3,2025-12-19 00:33:31,0.6.4
+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 HBM3,2025-12-19 00:33:31,0.6.4
+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 HBM3,2025-12-19 00:33:41,0.6.4
+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 HBM3,2025-12-19 00:33:41,0.6.4
+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 HBM3,2025-12-19 00:33:41,0.6.4
+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 HBM3,2025-12-19 00:33:41,0.6.4
+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 HBM3,2025-12-19 00:33:41,0.6.4
+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 HBM3,2025-12-19 00:33:41,0.6.4
+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 HBM3,2025-12-19 00:33:41,0.6.4
+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 HBM3,2025-12-19 00:33:41,0.6.4
+fused_add_rms_norm,liger_fused_add_rms_norm,forward,speed,ms,H,hidden size,1024,0.019999999552965164,0.018880000337958336,0.020767999812960625,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:50,0.6.4
+fused_add_rms_norm,liger_fused_add_rms_norm,forward,speed,ms,H,hidden size,2048,0.028672000393271446,0.028416000306606293,0.029343999922275543,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:50,0.6.4
+fused_add_rms_norm,liger_fused_add_rms_norm,forward,speed,ms,H,hidden size,4096,0.050944000482559204,0.050303999334573746,0.05161599814891815,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:50,0.6.4
+fused_add_rms_norm,liger_fused_add_rms_norm,forward,speed,ms,H,hidden size,8192,0.09616000205278397,0.0958079993724823,0.09651199728250504,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:50,0.6.4
+fused_add_rms_norm,liger_fused_add_rms_norm,forward,speed,ms,H,hidden size,16384,0.185248002409935,0.1847040057182312,0.18569600582122803,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:50,0.6.4
+fused_add_rms_norm,liger_fused_add_rms_norm,forward,speed,ms,H,hidden size,32768,0.43961599469184875,0.43887999653816223,0.44041600823402405,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:50,0.6.4
+fused_add_rms_norm,huggingface,forward,speed,ms,H,hidden size,1024,0.04800000041723251,0.04742399975657463,0.048448000103235245,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:53,0.6.4
+fused_add_rms_norm,huggingface,forward,speed,ms,H,hidden size,2048,0.07462400197982788,0.07407999783754349,0.07529599964618683,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:53,0.6.4
+fused_add_rms_norm,huggingface,forward,speed,ms,H,hidden size,4096,0.15308800339698792,0.15238399803638458,0.15374079644680022,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:53,0.6.4
+fused_add_rms_norm,huggingface,forward,speed,ms,H,hidden size,8192,0.275519996881485,0.27484801411628723,0.2761920094490051,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:53,0.6.4
+fused_add_rms_norm,huggingface,forward,speed,ms,H,hidden size,16384,0.5171200037002563,0.516319990158081,0.5178560018539429,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:53,0.6.4
+fused_add_rms_norm,huggingface,forward,speed,ms,H,hidden size,32768,0.9955199956893921,0.9947519898414612,0.9962239861488342,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:53,0.6.4
+fused_add_rms_norm,liger_rms_norm,forward,speed,ms,H,hidden size,1024,0.0379519984126091,0.03670400008559227,0.039712000638246536,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:57,0.6.4
+fused_add_rms_norm,liger_rms_norm,forward,speed,ms,H,hidden size,2048,0.049247998744249344,0.0488319993019104,0.049855999648571014,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:57,0.6.4
+fused_add_rms_norm,liger_rms_norm,forward,speed,ms,H,hidden size,4096,0.09788800030946732,0.09747199714183807,0.09827200323343277,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:57,0.6.4
+fused_add_rms_norm,liger_rms_norm,forward,speed,ms,H,hidden size,8192,0.18092800676822662,0.18025599420070648,0.18140800297260284,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:57,0.6.4
+fused_add_rms_norm,liger_rms_norm,forward,speed,ms,H,hidden size,16384,0.3452480137348175,0.34467199444770813,0.34598401188850403,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:57,0.6.4
+fused_add_rms_norm,liger_rms_norm,forward,speed,ms,H,hidden size,32768,0.6922240257263184,0.6914880275726318,0.6929600238800049,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:33:57,0.6.4
+fused_add_rms_norm,liger_fused_add_rms_norm,full,speed,ms,H,hidden size,1024,0.26953598856925964,0.25626240372657777,0.28940799832344055,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:02,0.6.4
+fused_add_rms_norm,liger_fused_add_rms_norm,full,speed,ms,H,hidden size,2048,0.2675680071115494,0.25578240752220155,0.28513280749320985,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:02,0.6.4
+fused_add_rms_norm,liger_fused_add_rms_norm,full,speed,ms,H,hidden size,4096,0.26532799005508423,0.2542400062084198,0.28172799944877625,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:02,0.6.4
+fused_add_rms_norm,liger_fused_add_rms_norm,full,speed,ms,H,hidden size,8192,0.27192001044750214,0.26102399826049805,0.287200003862381,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:02,0.6.4
+fused_add_rms_norm,liger_fused_add_rms_norm,full,speed,ms,H,hidden size,16384,1.2831679582595825,1.2811839580535889,1.2849215745925904,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:02,0.6.4
+fused_add_rms_norm,liger_fused_add_rms_norm,full,speed,ms,H,hidden size,32768,1.9645119905471802,1.9608383893966674,1.9686911821365356,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:02,0.6.4
+fused_add_rms_norm,huggingface,full,speed,ms,H,hidden size,1024,0.24513600021600723,0.23836800456047058,0.25397120118141175,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:05,0.6.4
+fused_add_rms_norm,huggingface,full,speed,ms,H,hidden size,2048,0.28934401273727417,0.28748801350593567,0.29123198986053467,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:05,0.6.4
+fused_add_rms_norm,huggingface,full,speed,ms,H,hidden size,4096,0.5511040091514587,0.5489152193069458,0.5532479882240295,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:05,0.6.4
+fused_add_rms_norm,huggingface,full,speed,ms,H,hidden size,8192,1.0046080350875854,1.0024000406265259,1.0067520141601562,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:05,0.6.4
+fused_add_rms_norm,huggingface,full,speed,ms,H,hidden size,16384,1.9005119800567627,1.8979840278625488,1.9028160572052002,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:05,0.6.4
+fused_add_rms_norm,huggingface,full,speed,ms,H,hidden size,32768,3.6795520782470703,3.6771841049194336,3.681823968887329,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:05,0.6.4
+fused_add_rms_norm,liger_rms_norm,full,speed,ms,H,hidden size,1024,0.3606400042772293,0.3477120101451874,0.3798015892505646,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:09,0.6.4
+fused_add_rms_norm,liger_rms_norm,full,speed,ms,H,hidden size,2048,0.3598400056362152,0.34835838675498965,0.37509760856628416,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:09,0.6.4
+fused_add_rms_norm,liger_rms_norm,full,speed,ms,H,hidden size,4096,0.35600000619888306,0.34384000301361084,0.371616005897522,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:09,0.6.4
+fused_add_rms_norm,liger_rms_norm,full,speed,ms,H,hidden size,8192,0.5687360167503357,0.5668799877166748,0.570464015007019,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:09,0.6.4
+fused_add_rms_norm,liger_rms_norm,full,speed,ms,H,hidden size,16384,1.119488000869751,1.1171136140823363,1.1215807676315308,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:09,0.6.4
+fused_add_rms_norm,liger_rms_norm,full,speed,ms,H,hidden size,32768,2.7628159523010254,2.759065675735474,2.7676416397094727,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:09,0.6.4
+fused_add_rms_norm,liger_fused_add_rms_norm,backward,speed,ms,H,hidden size,1024,0.1334879994392395,0.12656640112400055,0.1409855991601944,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:12,0.6.4
+fused_add_rms_norm,liger_fused_add_rms_norm,backward,speed,ms,H,hidden size,2048,0.13344000279903412,0.12660479843616484,0.14035199582576752,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:12,0.6.4
+fused_add_rms_norm,liger_fused_add_rms_norm,backward,speed,ms,H,hidden size,4096,0.1340160071849823,0.12636800110340118,0.1420159935951233,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:12,0.6.4
+fused_add_rms_norm,liger_fused_add_rms_norm,backward,speed,ms,H,hidden size,8192,0.15676799416542053,0.15561600029468536,0.15798400342464447,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:12,0.6.4
+fused_add_rms_norm,liger_fused_add_rms_norm,backward,speed,ms,H,hidden size,16384,1.102112054824829,1.100383996963501,1.1038336038589478,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:12,0.6.4
+fused_add_rms_norm,liger_fused_add_rms_norm,backward,speed,ms,H,hidden size,32768,1.533728003501892,1.5300159454345703,1.5372480154037476,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:12,0.6.4
+fused_add_rms_norm,huggingface,backward,speed,ms,H,hidden size,1024,0.15220800042152405,0.1441279947757721,0.1599104017019272,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:14,0.6.4
+fused_add_rms_norm,huggingface,backward,speed,ms,H,hidden size,2048,0.21491199731826782,0.21420800685882568,0.21564799547195435,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:14,0.6.4
+fused_add_rms_norm,huggingface,backward,speed,ms,H,hidden size,4096,0.400191992521286,0.3993920087814331,0.4010559916496277,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:14,0.6.4
+fused_add_rms_norm,huggingface,backward,speed,ms,H,hidden size,8192,0.7319679856300354,0.7309439778327942,0.7329599857330322,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:14,0.6.4
+fused_add_rms_norm,huggingface,backward,speed,ms,H,hidden size,16384,1.3854560256004333,1.3841919898986816,1.386687994003296,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:14,0.6.4
+fused_add_rms_norm,huggingface,backward,speed,ms,H,hidden size,32768,2.687567949295044,2.6858688831329345,2.6885823249816894,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:14,0.6.4
+fused_add_rms_norm,liger_rms_norm,backward,speed,ms,H,hidden size,1024,0.18223999440670013,0.17452800273895264,0.1900160014629364,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:17,0.6.4
+fused_add_rms_norm,liger_rms_norm,backward,speed,ms,H,hidden size,2048,0.18092800676822662,0.17369599640369415,0.18911999464035034,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:17,0.6.4
+fused_add_rms_norm,liger_rms_norm,backward,speed,ms,H,hidden size,4096,0.21571199595928192,0.21449600160121918,0.21721599996089935,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:17,0.6.4
+fused_add_rms_norm,liger_rms_norm,backward,speed,ms,H,hidden size,8192,0.3924799859523773,0.39103999733924866,0.39388159513473514,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:17,0.6.4
+fused_add_rms_norm,liger_rms_norm,backward,speed,ms,H,hidden size,16384,0.7791360020637512,0.7772799730300903,0.7810559868812561,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:17,0.6.4
+fused_add_rms_norm,liger_rms_norm,backward,speed,ms,H,hidden size,32768,2.0764319896698,2.0732352256774904,2.0795839309692385,"{""M"": 2048, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:17,0.6.4
+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 HBM3,2025-12-19 00:34:17,0.6.4
+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 HBM3,2025-12-19 00:34:17,0.6.4
+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 HBM3,2025-12-19 00:34:17,0.6.4
+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 HBM3,2025-12-19 00:34:17,0.6.4
+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 HBM3,2025-12-19 00:34:17,0.6.4
+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 HBM3,2025-12-19 00:34:17,0.6.4
+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 HBM3,2025-12-19 00:34:17,0.6.4
+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 HBM3,2025-12-19 00:34:17,0.6.4
+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 HBM3,2025-12-19 00:34:17,0.6.4
+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 HBM3,2025-12-19 00:34:17,0.6.4
+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 HBM3,2025-12-19 00:34:17,0.6.4
+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 HBM3,2025-12-19 00:34:17,0.6.4
+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 HBM3,2025-12-19 00:34:18,0.6.4
+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 HBM3,2025-12-19 00:34:18,0.6.4
+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 HBM3,2025-12-19 00:34:18,0.6.4
+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 HBM3,2025-12-19 00:34:18,0.6.4
+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 HBM3,2025-12-19 00:34:18,0.6.4
+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 HBM3,2025-12-19 00:34:18,0.6.4
+fused_linear_cross_entropy,liger,forward,speed,ms,BT,B x T,4096,156.69088745117188,156.69088745117188,156.69088745117188,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:40,0.6.4
+fused_linear_cross_entropy,liger,forward,speed,ms,BT,B x T,8192,161.38800048828125,161.38800048828125,161.38800048828125,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:40,0.6.4
+fused_linear_cross_entropy,liger,forward,speed,ms,BT,B x T,16384,194.7716827392578,194.7716827392578,194.7716827392578,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:40,0.6.4
+fused_linear_cross_entropy,liger,forward,speed,ms,BT,B x T,32768,274.4150390625,274.4150390625,274.4150390625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:40,0.6.4
+fused_linear_cross_entropy,liger-fp32-accum,forward,speed,ms,BT,B x T,4096,158.42201232910156,158.42201232910156,158.42201232910156,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:57,0.6.4
+fused_linear_cross_entropy,liger-fp32-accum,forward,speed,ms,BT,B x T,8192,162.74441528320312,162.74441528320312,162.74441528320312,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:57,0.6.4
+fused_linear_cross_entropy,liger-fp32-accum,forward,speed,ms,BT,B x T,16384,194.73992919921875,194.73992919921875,194.73992919921875,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:57,0.6.4
+fused_linear_cross_entropy,liger-fp32-accum,forward,speed,ms,BT,B x T,32768,268.8746643066406,268.8746643066406,268.8746643066406,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:34:57,0.6.4
+fused_linear_cross_entropy,huggingface,forward,speed,ms,BT,B x T,4096,6.8304479122161865,6.771084785461426,7.140422344207764,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:35:10,0.6.4
+fused_linear_cross_entropy,huggingface,forward,speed,ms,BT,B x T,8192,13.322336196899414,13.157375717163086,13.481248092651366,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:35:10,0.6.4
+fused_linear_cross_entropy,huggingface,forward,speed,ms,BT,B x T,16384,27.578975677490234,27.513023376464844,28.122470474243165,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:35:10,0.6.4
+fused_linear_cross_entropy,huggingface,forward,speed,ms,BT,B x T,32768,54.94220733642578,54.94220733642578,54.94220733642578,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:35:10,0.6.4
+fused_linear_cross_entropy,liger,backward,speed,ms,BT,B x T,4096,1.1142079830169678,1.110924768447876,1.1197247743606566,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:35:22,0.6.4
+fused_linear_cross_entropy,liger,backward,speed,ms,BT,B x T,8192,1.1333119869232178,1.132262372970581,1.1351040363311768,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:35:22,0.6.4
+fused_linear_cross_entropy,liger,backward,speed,ms,BT,B x T,16384,1.1767679452896118,1.175276756286621,1.1779711961746215,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:35:22,0.6.4
+fused_linear_cross_entropy,liger,backward,speed,ms,BT,B x T,32768,1.2665599584579468,1.2654847860336305,1.2681344032287598,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:35:22,0.6.4
+fused_linear_cross_entropy,liger-fp32-accum,backward,speed,ms,BT,B x T,4096,1.114143967628479,1.1120384216308594,1.11802237033844,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:35:35,0.6.4
+fused_linear_cross_entropy,liger-fp32-accum,backward,speed,ms,BT,B x T,8192,1.1355520486831665,1.133075213432312,1.1413823843002318,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:35:35,0.6.4
+fused_linear_cross_entropy,liger-fp32-accum,backward,speed,ms,BT,B x T,16384,1.183135986328125,1.1813376426696778,1.1880639791488647,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:35:35,0.6.4
+fused_linear_cross_entropy,liger-fp32-accum,backward,speed,ms,BT,B x T,32768,1.266592025756836,1.2656255960464478,1.2679168224334716,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:35:35,0.6.4
+fused_linear_cross_entropy,huggingface,backward,speed,ms,BT,B x T,4096,13.218303680419922,13.216563034057618,14.318015670776369,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:35:48,0.6.4
+fused_linear_cross_entropy,huggingface,backward,speed,ms,BT,B x T,8192,26.221952438354492,25.904288482666015,27.34290580749512,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:35:48,0.6.4
+fused_linear_cross_entropy,huggingface,backward,speed,ms,BT,B x T,16384,53.284095764160156,53.284095764160156,53.284095764160156,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:35:48,0.6.4
+fused_linear_cross_entropy,huggingface,backward,speed,ms,BT,B x T,32768,127.34575653076172,127.34575653076172,127.34575653076172,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:35:48,0.6.4
+fused_linear_cross_entropy,liger,full,speed,ms,BT,B x T,4096,157.5265350341797,157.5265350341797,157.5265350341797,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:36:05,0.6.4
+fused_linear_cross_entropy,liger,full,speed,ms,BT,B x T,8192,162.66957092285156,162.66957092285156,162.66957092285156,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:36:05,0.6.4
+fused_linear_cross_entropy,liger,full,speed,ms,BT,B x T,16384,195.0821075439453,195.0821075439453,195.0821075439453,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:36:05,0.6.4
+fused_linear_cross_entropy,liger,full,speed,ms,BT,B x T,32768,272.62078857421875,272.62078857421875,272.62078857421875,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:36:05,0.6.4
+fused_linear_cross_entropy,liger-fp32-accum,full,speed,ms,BT,B x T,4096,159.4628448486328,159.4628448486328,159.4628448486328,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:36:23,0.6.4
+fused_linear_cross_entropy,liger-fp32-accum,full,speed,ms,BT,B x T,8192,163.8953857421875,163.8953857421875,163.8953857421875,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:36:23,0.6.4
+fused_linear_cross_entropy,liger-fp32-accum,full,speed,ms,BT,B x T,16384,196.03549194335938,196.03549194335938,196.03549194335938,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:36:23,0.6.4
+fused_linear_cross_entropy,liger-fp32-accum,full,speed,ms,BT,B x T,32768,268.8150939941406,268.8150939941406,268.8150939941406,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:36:23,0.6.4
+fused_linear_cross_entropy,huggingface,full,speed,ms,BT,B x T,4096,21.57539176940918,21.4820858001709,21.63191032409668,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:36:37,0.6.4
+fused_linear_cross_entropy,huggingface,full,speed,ms,BT,B x T,8192,40.559791564941406,40.20396728515625,40.91561584472656,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:36:37,0.6.4
+fused_linear_cross_entropy,huggingface,full,speed,ms,BT,B x T,16384,81.76525115966797,81.76525115966797,81.76525115966797,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:36:37,0.6.4
+fused_linear_cross_entropy,huggingface,full,speed,ms,BT,B x T,32768,166.08128356933594,166.08128356933594,166.08128356933594,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:36:37,0.6.4
+fused_linear_cross_entropy,liger,no-grad-forward,speed,ms,BT,B x T,4096,156.77590942382812,156.77590942382812,156.77590942382812,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:36:54,0.6.4
+fused_linear_cross_entropy,liger,no-grad-forward,speed,ms,BT,B x T,8192,161.52476501464844,161.52476501464844,161.52476501464844,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:36:54,0.6.4
+fused_linear_cross_entropy,liger,no-grad-forward,speed,ms,BT,B x T,16384,195.47900390625,195.47900390625,195.47900390625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:36:54,0.6.4
+fused_linear_cross_entropy,liger,no-grad-forward,speed,ms,BT,B x T,32768,273.3216552734375,273.3216552734375,273.3216552734375,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:36:54,0.6.4
+fused_linear_cross_entropy,liger-fp32-accum,no-grad-forward,speed,ms,BT,B x T,4096,158.3824920654297,158.3824920654297,158.3824920654297,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:37:12,0.6.4
+fused_linear_cross_entropy,liger-fp32-accum,no-grad-forward,speed,ms,BT,B x T,8192,162.75485229492188,162.75485229492188,162.75485229492188,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:37:12,0.6.4
+fused_linear_cross_entropy,liger-fp32-accum,no-grad-forward,speed,ms,BT,B x T,16384,194.67295837402344,194.67295837402344,194.67295837402344,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:37:12,0.6.4
+fused_linear_cross_entropy,liger-fp32-accum,no-grad-forward,speed,ms,BT,B x T,32768,268.234375,268.234375,268.234375,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:37:12,0.6.4
+fused_linear_cross_entropy,huggingface,no-grad-forward,speed,ms,BT,B x T,4096,7.028271913528442,6.876454448699951,7.0309953689575195,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:37:24,0.6.4
+fused_linear_cross_entropy,huggingface,no-grad-forward,speed,ms,BT,B x T,8192,13.322591781616211,13.263999938964844,13.423558807373047,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:37:24,0.6.4
+fused_linear_cross_entropy,huggingface,no-grad-forward,speed,ms,BT,B x T,16384,28.00102424621582,27.880275344848634,28.085446166992188,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:37:24,0.6.4
+fused_linear_cross_entropy,huggingface,no-grad-forward,speed,ms,BT,B x T,32768,55.22470474243164,55.22470474243164,55.22470474243164,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:37:24,0.6.4
+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 HBM3,2025-12-19 00:37:43,0.6.4
+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 HBM3,2025-12-19 00:37:43,0.6.4
+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 HBM3,2025-12-19 00:37:43,0.6.4
+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 HBM3,2025-12-19 00:37:43,0.6.4
+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 HBM3,2025-12-19 00:38:02,0.6.4
+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 HBM3,2025-12-19 00:38:02,0.6.4
+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 HBM3,2025-12-19 00:38:02,0.6.4
+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 HBM3,2025-12-19 00:38:02,0.6.4
+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 HBM3,2025-12-19 00:38:14,0.6.4
+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 HBM3,2025-12-19 00:38:14,0.6.4
+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 HBM3,2025-12-19 00:38:14,0.6.4
+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 HBM3,2025-12-19 00:38:14,0.6.4
+fused_linear_jsd,liger,forward,speed,ms,BT,B x T,1024,128.8693084716797,128.8693084716797,128.8693084716797,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:32,0.6.4
+fused_linear_jsd,liger,forward,speed,ms,BT,B x T,2048,150.79376220703125,150.79376220703125,150.79376220703125,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:32,0.6.4
+fused_linear_jsd,liger,forward,speed,ms,BT,B x T,4096,176.36834716796875,176.36834716796875,176.36834716796875,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:32,0.6.4
+fused_linear_jsd,liger,forward,speed,ms,BT,B x T,8192,225.6640625,225.6640625,225.6640625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:32,0.6.4
+fused_linear_jsd,torch,forward,speed,ms,BT,B x T,1024,9.594031810760498,9.591072082519531,9.597657203674316,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:34,0.6.4
+fused_linear_jsd,torch,forward,speed,ms,BT,B x T,2048,18.897504806518555,18.888570022583007,18.909984588623047,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:34,0.6.4
+fused_linear_jsd,torch,forward,speed,ms,BT,B x T,4096,37.6517276763916,37.64956817626953,37.65388717651367,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:34,0.6.4
+fused_linear_jsd,torch,forward,speed,ms,BT,B x T,8192,74.84070587158203,74.84070587158203,74.84070587158203,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:34,0.6.4
+fused_linear_jsd,liger,backward,speed,ms,BT,B x T,1024,1.0938880443572998,1.0903359651565552,1.111583948135376,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:35,0.6.4
+fused_linear_jsd,liger,backward,speed,ms,BT,B x T,2048,1.1016000509262085,1.0947840213775635,1.1163647890090942,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:35,0.6.4
+fused_linear_jsd,liger,backward,speed,ms,BT,B x T,4096,1.1088480353355408,1.1046720504760743,1.1278528213500978,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:35,0.6.4
+fused_linear_jsd,liger,backward,speed,ms,BT,B x T,8192,1.1303040385246277,1.1263680458068848,1.143167996406555,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:35,0.6.4
+fused_linear_jsd,torch,backward,speed,ms,BT,B x T,1024,14.610912322998047,14.609523010253907,14.613804435729982,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:37,0.6.4
+fused_linear_jsd,torch,backward,speed,ms,BT,B x T,2048,28.85081672668457,28.848415756225588,28.85682601928711,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:37,0.6.4
+fused_linear_jsd,torch,backward,speed,ms,BT,B x T,4096,57.43734359741211,57.43734359741211,57.43734359741211,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:37,0.6.4
+fused_linear_jsd,torch,backward,speed,ms,BT,B x T,8192,116.63577270507812,116.63577270507812,116.63577270507812,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:37,0.6.4
+fused_linear_jsd,liger,full,speed,ms,BT,B x T,1024,130.15577697753906,130.15577697753906,130.15577697753906,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:42,0.6.4
+fused_linear_jsd,liger,full,speed,ms,BT,B x T,2048,151.910400390625,151.910400390625,151.910400390625,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:42,0.6.4
+fused_linear_jsd,liger,full,speed,ms,BT,B x T,4096,177.51878356933594,177.51878356933594,177.51878356933594,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:42,0.6.4
+fused_linear_jsd,liger,full,speed,ms,BT,B x T,8192,226.9738311767578,226.9738311767578,226.9738311767578,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:42,0.6.4
+fused_linear_jsd,torch,full,speed,ms,BT,B x T,1024,26.235679626464844,26.23514175415039,26.240153121948243,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:46,0.6.4
+fused_linear_jsd,torch,full,speed,ms,BT,B x T,2048,49.826446533203125,49.8063346862793,49.84655838012695,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:46,0.6.4
+fused_linear_jsd,torch,full,speed,ms,BT,B x T,4096,97.57353973388672,97.57353973388672,97.57353973388672,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:46,0.6.4
+fused_linear_jsd,torch,full,speed,ms,BT,B x T,8192,197.06179809570312,197.06179809570312,197.06179809570312,"{""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:38:46,0.6.4
+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 HBM3,2025-12-19 00:38:52,0.6.4
+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 HBM3,2025-12-19 00:38:52,0.6.4
+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 HBM3,2025-12-19 00:38:52,0.6.4
+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 HBM3,2025-12-19 00:38:52,0.6.4
+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 HBM3,2025-12-19 00:38:54,0.6.4
+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 HBM3,2025-12-19 00:38:54,0.6.4
+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 HBM3,2025-12-19 00:38:54,0.6.4
+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 HBM3,2025-12-19 00:38:54,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,64,0.3012799918651581,0.29625600576400757,0.30887681245803833,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:06,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,128,0.29363200068473816,0.2888511955738068,0.29980159401893614,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:06,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,256,0.2951679974794388,0.29040640592575073,0.302649587392807,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:06,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,512,0.29715199768543243,0.29286399483680725,0.3035776078701019,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:06,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,1024,0.31140799820423126,0.3085759937763214,0.31324800848960876,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:06,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,2048,0.788640022277832,0.7875135898590088,0.7895424008369445,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:06,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,4096,2.6140799522399902,2.612511920928955,2.6166463851928707,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:06,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,64,5.824368000030518,5.810272216796875,5.835328102111816,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:13,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,128,11.57809591293335,11.54837760925293,11.587110328674317,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:13,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,256,22.954992294311523,22.934368133544922,22.97122573852539,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:13,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,512,45.95975875854492,45.95723419189453,45.962283325195315,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:13,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,1024,91.66575622558594,91.66575622558594,91.66575622558594,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:13,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,2048,182.9102020263672,182.9102020263672,182.9102020263672,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:13,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,4096,368.0877380371094,368.0877380371094,368.0877380371094,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:13,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,64,1.3530879616737366,1.30033278465271,1.4973631858825684,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:23,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,128,1.2507519721984863,1.2269376039505004,1.2894336462020874,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:23,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,256,1.2379040122032166,1.212998366355896,1.3064640045166018,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:23,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,512,1.2917759418487549,1.270080041885376,1.3279168128967287,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:23,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,1024,1.3218719959259033,1.2925824403762818,1.439174437522888,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:23,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,2048,2.698240041732788,2.6935231685638428,2.705401563644409,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:23,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,4096,8.710304260253906,8.705280303955078,8.775712013244629,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:23,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,64,7.1281280517578125,7.1093439102172855,7.144921398162841,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:31,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,128,13.279520034790039,13.260115242004394,13.437369918823244,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:31,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,256,25.524127960205078,25.474957275390626,25.53403511047363,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:31,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,512,50.2276496887207,50.2276496887207,50.2276496887207,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:31,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,1024,101.00041961669922,101.00041961669922,101.00041961669922,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:31,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,2048,199.40521240234375,199.40521240234375,199.40521240234375,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:31,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,4096,403.5535583496094,403.5535583496094,403.5535583496094,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:31,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,64,0.7002879977226257,0.6849407911300659,0.7379391908645629,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:32,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,128,0.6916000247001648,0.6719487905502319,0.7107328176498413,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:32,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,256,0.7047039866447449,0.6915136218070983,0.733900821208954,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:32,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,512,0.743008017539978,0.7292991995811463,0.762489628791809,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:32,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,1024,0.7464640140533447,0.741433584690094,0.7488960027694702,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:32,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,2048,1.9245760440826416,1.9002816438674928,1.9286784172058105,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:32,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,4096,6.167168140411377,6.094790554046631,6.174803352355957,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:32,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,64,0.5751680135726929,0.5678719878196716,0.6041280031204224,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:34,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,128,0.5765439867973328,0.5661120057106017,0.6020736217498779,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:34,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,256,0.5957760214805603,0.5871679782867432,0.6113920211791992,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:34,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,512,0.6178240180015564,0.6072256088256835,0.6359039902687074,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:34,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,1024,0.9600159823894501,0.9582783818244934,0.9618175983428956,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:34,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,2048,2.5819520950317383,2.579974365234375,2.5831360340118406,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:34,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,4096,8.732192039489746,8.731136322021484,8.73459243774414,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:34,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,64,0.310016006231308,0.304640007019043,0.3169663965702057,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:39,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,128,0.3107520043849945,0.30588799715042114,0.31510400772094727,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:39,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,256,0.31595200300216675,0.3111871898174286,0.3229695975780487,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:39,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,512,0.43091198801994324,0.43001601099967957,0.43182079792022704,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:39,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,1024,0.9967359900474548,0.9945600032806396,0.9978752255439758,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:39,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,2048,2.7537600994110107,2.7275968551635743,2.755372714996338,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:39,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,4096,8.541376113891602,8.482879638671875,8.542240142822266,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:39,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,64,5.8352320194244385,5.816959857940674,5.849440097808838,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:47,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,128,12.430895805358887,12.39824619293213,12.467289352416993,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:47,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,256,24.684096336364746,24.558956909179685,24.793709182739256,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:47,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,512,49.63443183898926,49.46631698608399,49.802546691894534,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:47,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,1024,99.96943664550781,99.96943664550781,99.96943664550781,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:47,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,2048,200.44912719726562,200.44912719726562,200.44912719726562,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:47,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,4096,408.2440185546875,408.2440185546875,408.2440185546875,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:47,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,64,1.2873120307922363,1.26627836227417,1.3381440401077271,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:56,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,128,1.288256049156189,1.26627197265625,1.3691583633422852,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:56,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,256,1.3516160249710083,1.3213759660720825,1.4502400159835815,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:56,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,512,1.3902720212936401,1.3847424030303954,1.3942464351654054,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:56,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,1024,3.1851840019226074,3.1431552410125736,3.189964771270752,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:56,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,2048,8.708160400390625,8.706080436706543,8.71622371673584,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:56,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,4096,27.727712631225586,27.620959854125978,27.732723999023438,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:39:56,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,64,7.131999969482422,7.110783958435059,7.155340862274169,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:04,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,128,13.307744026184082,13.257088088989258,13.35245418548584,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:04,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,256,25.557151794433594,25.546706771850587,25.690015411376955,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:04,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,512,51.02326583862305,51.02326583862305,51.02326583862305,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:04,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,1024,102.33586883544922,102.33586883544922,102.33586883544922,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:04,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,2048,207.75875854492188,207.75875854492188,207.75875854492188,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:04,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,4096,410.4659118652344,410.4659118652344,410.4659118652344,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:04,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,64,0.650879979133606,0.6369600296020508,0.6610879898071289,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:05,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,128,0.6570880115032196,0.6435136079788208,0.692076814174652,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:05,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,256,0.6955199837684631,0.6800960063934326,0.7267584085464478,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:05,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,512,0.9535999894142151,0.9523711919784545,0.9553279876708984,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:05,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,1024,2.1703039407730103,2.1677184104919434,2.1726528644561767,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:05,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,2048,6.04584002494812,6.044191837310791,6.04911994934082,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:05,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,4096,19.266624450683594,19.237459564208983,19.269497299194335,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:05,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,64,0.5848640203475952,0.5661504030227661,0.6164544105529786,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:08,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,128,0.5734080076217651,0.5470271944999695,0.6092095971107483,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:08,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,256,0.574400007724762,0.5638719797134399,0.6119679808616638,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:08,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,512,1.0858399868011475,1.0846271991729737,1.0873216152191163,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:08,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,1024,2.683199882507324,2.6813440322875977,2.6846463680267334,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:08,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,2048,7.912480115890503,7.910950374603272,7.918918609619141,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:08,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,4096,26.621440887451172,26.62082633972168,26.621651458740235,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:08,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,64,0.3259039968252182,0.3226880133152008,0.3303680121898651,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:13,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,128,0.30905601382255554,0.30556800961494446,0.3130112111568451,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:13,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,256,0.31409600377082825,0.3099839985370636,0.3192319869995117,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:13,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,512,0.35574400424957275,0.3550400137901306,0.3566719889640808,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:13,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,1024,0.7106559872627258,0.7078335762023926,0.7120960235595704,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:13,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,2048,1.8683520555496216,1.857375979423523,1.8703680038452148,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:13,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,4096,6.0370399951934814,6.033952236175537,6.06387186050415,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:13,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,64,7.923775911331177,7.883551979064941,7.951385784149171,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:23,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,128,15.80683183670044,15.785216331481934,15.833919525146484,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:23,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,256,31.617855072021484,31.538597869873048,31.620851135253908,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:23,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,512,62.826560974121094,62.826560974121094,62.826560974121094,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:23,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,1024,126.36505889892578,126.36505889892578,126.36505889892578,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:23,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,2048,253.8020782470703,253.8020782470703,253.8020782470703,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:23,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,4096,481.2433776855469,481.2433776855469,481.2433776855469,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:23,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,64,1.3175040483474731,1.2882944345474243,1.3756927728652955,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:32,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,128,1.263808012008667,1.2403584241867065,1.3103679895401,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:32,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,256,1.2727359533309937,1.2508288145065307,1.2944767475128174,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:32,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,512,1.3378239870071411,1.315168023109436,1.4660799503326416,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:32,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,1024,2.378416061401367,2.3564608097076416,2.383647918701172,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:32,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,2048,6.1721601486206055,6.15393295288086,6.175468826293946,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:32,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,4096,19.518815994262695,19.49176940917969,19.524077224731446,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:32,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,64,8.794415950775146,8.74684829711914,8.92148494720459,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:42,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,128,16.64998435974121,16.588851928710938,16.734822845458986,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:42,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,256,32.450592041015625,32.4291275024414,32.46380081176758,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:42,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,512,65.71001434326172,65.71001434326172,65.71001434326172,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:42,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,1024,128.66796875,128.66796875,128.66796875,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:42,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,2048,260.10223388671875,260.10223388671875,260.10223388671875,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:42,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,4096,532.3633422851562,532.3633422851562,532.3633422851562,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:42,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,64,0.6933920085430145,0.6808575868606568,0.7192896127700805,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:43,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,128,0.7092799842357635,0.6875200271606445,0.7427200078964233,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:43,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,256,0.6918080151081085,0.6786239981651306,0.7157759904861449,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:43,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,512,0.8234879970550537,0.8215680122375488,0.8247936129570007,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:43,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,1024,1.6711360216140747,1.6678719520568848,1.6752959489822388,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:43,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,2048,4.324399948120117,4.258879852294922,4.328851413726807,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:43,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,4096,13.425503730773926,13.419244575500487,13.427212524414063,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:43,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,64,0.5675840079784393,0.5592511892318726,0.5767103791236877,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:46,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,128,0.5731520056724548,0.5630080103874207,0.5879999995231628,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:46,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,256,0.5730240046977997,0.564364778995514,0.588812792301178,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:46,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,512,0.9217120110988617,0.9200448036193849,0.9238143801689149,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:46,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,1024,1.9833600521087646,1.9818047761917115,1.9851392030715942,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:46,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,2048,5.568511962890625,5.56561918258667,5.571609687805176,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:46,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,4096,18.520095825195312,18.51799659729004,18.522022247314453,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:46,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,64,0.297791987657547,0.29456639289855957,0.3019584059715271,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:50,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,128,0.29289600253105164,0.28911359906196593,0.29703680276870725,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:50,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,256,0.2948960065841675,0.291456013917923,0.3004159927368164,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:50,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,512,0.302623987197876,0.2978239953517914,0.30768001079559326,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:50,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,1024,0.31836798787117004,0.31436800956726074,0.3241087973117828,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:50,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,2048,0.7906399965286255,0.7824959993362426,0.7919039726257324,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:50,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,4096,2.6201759576797485,2.618175983428955,2.630431890487671,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:50,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,64,5.844671964645386,5.820703983306885,5.876992225646973,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:57,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,128,11.810816287994385,11.702201271057127,12.051039886474609,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:57,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,256,23.273280143737793,23.24299507141113,23.37111053466797,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:57,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,512,46.869184494018555,46.86213836669922,46.876230621337896,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:57,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,1024,100.3872299194336,100.3872299194336,100.3872299194336,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:57,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,2048,199.62413024902344,199.62413024902344,199.62413024902344,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:57,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,4096,378.3080139160156,378.3080139160156,378.3080139160156,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:57,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,64,1.319808006286621,1.2829439640045166,1.412768006324768,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:58,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,128,1.2710400223731995,1.2447615623474122,1.3243840217590335,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:58,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,256,1.2761439681053162,1.2525440454483032,1.338752031326294,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:58,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,512,1.3256640434265137,1.3019136190414429,1.36494722366333,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:58,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,1024,1.3206560015678406,1.304428791999817,1.3391040086746215,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:58,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,2048,2.724191904067993,2.7213759422302246,2.726598358154297,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:58,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,4096,8.7193603515625,8.712096214294434,8.72054386138916,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:40:58,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,64,7.007728099822998,6.61592321395874,7.030534553527833,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:06,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,128,13.309743881225586,13.277510261535644,13.337094497680663,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:06,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,256,25.724191665649414,25.713996124267577,25.73588409423828,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:06,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,512,50.95868682861328,50.95868682861328,50.95868682861328,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:06,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,1024,100.78125,100.78125,100.78125,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:06,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,2048,202.91897583007812,202.91897583007812,202.91897583007812,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:06,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,4096,411.9269104003906,411.9269104003906,411.9269104003906,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:06,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,64,0.681119978427887,0.6701760292053223,0.6980159878730774,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:07,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,128,0.7009919881820679,0.6881600022315979,0.7401599884033203,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:07,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,256,0.6943359971046448,0.6821311712265015,0.7091263890266418,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:07,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,512,0.7316320240497589,0.718668794631958,0.7522176027297973,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:07,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,1024,0.7451039850711823,0.743232011795044,0.7471039891242981,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:07,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,2048,1.9322559833526611,1.9289599895477296,1.9348671674728393,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:07,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,4096,6.184415817260742,6.099647903442383,6.187180709838867,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:07,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,64,0.5655199885368347,0.5557760119438171,0.5779327988624573,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:10,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,128,0.560943990945816,0.5541120171546936,0.5683519840240479,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:10,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,256,0.5799200236797333,0.5704192042350769,0.600390386581421,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:10,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,512,0.6028800010681152,0.5935680270195007,0.6281920075416565,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:10,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,1024,0.9570879936218262,0.9562047839164733,0.9585471868515014,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:10,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,2048,2.5880000591278076,2.58599681854248,2.5900800228118896,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:10,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,4096,8.731424331665039,8.730208396911621,8.733951568603516,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:10,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,64,0.29022400081157684,0.28620800375938416,0.2956160008907318,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:15,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,128,0.30663999915122986,0.3024704039096832,0.3128384053707123,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:15,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,256,0.30665600299835205,0.30286720395088196,0.3137023985385895,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:15,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,512,0.3111039996147156,0.3063871920108795,0.3167872130870819,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:15,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,1024,0.31015999615192413,0.3055359899997711,0.31787520051002505,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:15,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,2048,0.33241599798202515,0.33113598823547363,0.33369600772857666,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:15,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,4096,1.1031999588012695,1.101094388961792,1.1063551664352418,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:15,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,64,6.232672214508057,6.21380500793457,6.269446277618408,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:22,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,128,12.353519916534424,12.337427139282227,12.379615783691406,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:22,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,256,24.507328033447266,24.498298263549806,24.521823501586915,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:22,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,512,49.13928031921387,49.073434448242196,49.205126190185545,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:22,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,1024,98.26675415039062,98.26675415039062,98.26675415039062,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:22,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,2048,196.1670379638672,196.1670379638672,196.1670379638672,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:22,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,4096,393.40655517578125,393.40655517578125,393.40655517578125,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:22,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,64,1.2451680302619934,1.2221440076828003,1.2729920148849487,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:30,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,128,1.2429920434951782,1.2266047954559327,1.290822434425354,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:30,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,256,1.2660000324249268,1.2361216068267822,1.5179200172424316,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:30,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,512,1.3516160249710083,1.2902400493621826,1.5024640560150146,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:30,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,1024,1.3409759998321533,1.3069376468658447,1.5200255632400514,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:30,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,2048,1.314560055732727,1.2903679847717284,1.41113600730896,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:30,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,4096,2.991007924079895,2.987359952926636,2.9935935974121097,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:30,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,64,7.07918381690979,7.056639862060547,7.348473453521729,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:37,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,128,12.490352153778076,12.413036727905274,12.687513732910157,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:37,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,256,24.282544136047363,24.016819381713866,25.39063720703125,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:37,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,512,50.39606475830078,50.39606475830078,50.39606475830078,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:37,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,1024,99.46454620361328,99.46454620361328,99.46454620361328,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:37,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,2048,198.44207763671875,198.44207763671875,198.44207763671875,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:37,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,4096,373.9899597167969,373.9899597167969,373.9899597167969,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:37,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,64,0.6785280108451843,0.651526403427124,0.7207744121551514,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:38,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,128,0.6782400012016296,0.6625728011131287,0.7056896090507507,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:38,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,256,0.6644480228424072,0.6482816100120544,0.6892352223396301,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:38,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,512,0.7072800099849701,0.6932608008384704,0.7323903799057008,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:38,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,1024,0.7079039812088013,0.6890175819396973,0.7538048028945923,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:38,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,2048,0.7025279998779297,0.6912960171699525,0.7142079949378968,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:38,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,4096,1.8841440081596375,1.8817408084869385,1.886195182800293,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:38,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,64,0.5567359924316406,0.5452415823936463,0.5802496194839477,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:40,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,128,0.5655840039253235,0.5484736204147338,0.6097024083137512,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:40,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,256,0.551967978477478,0.5425856113433838,0.5663936018943787,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:40,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,512,0.5879999995231628,0.5745216012001038,0.616044819355011,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:40,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,1024,0.5836319923400879,0.5755135893821717,0.6065408110618591,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:40,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,2048,1.0830719470977783,1.0815807819366454,1.0842623949050902,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:40,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,4096,3.6387200355529785,3.635103940963745,3.6425600051879883,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:40,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,64,0.2953920066356659,0.291129595041275,0.3007616102695465,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:45,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,128,0.3123199939727783,0.3072960078716278,0.31782400608062744,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:45,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,256,0.2945280075073242,0.29097598791122437,0.29919999837875366,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:45,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,512,0.3142080008983612,0.30916481018066405,0.31962879300117497,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:45,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,1024,0.32526399195194244,0.3202560007572174,0.3310847938060761,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:45,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,2048,0.9274240136146545,0.9246464014053344,0.9306240081787109,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:45,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,4096,3.323712110519409,3.319328022003174,3.33090558052063,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:45,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,64,5.820688009262085,5.809567928314209,5.84438419342041,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:52,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,128,11.61190414428711,11.580665588378906,11.641356658935546,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:52,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,256,23.030447959899902,23.015404891967773,23.053874969482422,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:52,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,512,45.83228874206543,45.812608337402345,45.851969146728514,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:52,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,1024,96.6941146850586,96.6941146850586,96.6941146850586,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:52,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,2048,198.2008056640625,198.2008056640625,198.2008056640625,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:52,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,4096,399.52593994140625,399.52593994140625,399.52593994140625,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:52,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,64,1.21041601896286,1.1891200542449951,1.2333439588546753,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:59,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,128,1.2240959405899048,1.1989568233489991,1.284127950668335,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:59,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,256,1.248703956604004,1.2147392272949218,1.3375359773635864,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:59,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,512,1.3042399883270264,1.2472512245178224,1.449740791320801,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:59,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,1024,1.2596960067749023,1.235200047492981,1.2902400493621826,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:59,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,2048,2.5964479446411133,2.595072031021118,2.60099196434021,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:59,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,4096,9.35368013381958,9.347705268859864,9.361362838745118,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:41:59,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,64,7.082719802856445,7.066784000396728,7.143238353729248,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:07,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,128,13.298239707946777,13.242713356018067,13.303769874572753,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:07,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,256,23.851760864257812,23.825568771362306,23.893440628051756,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:07,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,512,47.36924743652344,47.36924743652344,47.36924743652344,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:07,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,1024,93.13906860351562,93.13906860351562,93.13906860351562,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:07,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,2048,188.94720458984375,188.94720458984375,188.94720458984375,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:07,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,4096,411.7369384765625,411.7369384765625,411.7369384765625,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:07,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,64,0.6409279704093933,0.6316480040550232,0.6640959978103638,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:08,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,128,0.6341760158538818,0.6245184183120728,0.6459840059280395,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:08,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,256,0.6604320108890533,0.6428160071372986,0.701088011264801,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:08,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,512,0.6659520268440247,0.6528767943382263,0.67986558675766,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:08,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,1024,0.7046720087528229,0.6742720007896423,0.7432640194892883,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:08,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,2048,1.6770079731941223,1.6744320392608643,1.6794559955596924,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:08,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,4096,6.020175933837891,6.012832164764404,6.024767875671387,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:08,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,64,0.5160639882087708,0.5095360279083252,0.5211520195007324,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:10,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,128,0.5241280198097229,0.5146560072898865,0.5538367986679077,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:10,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,256,0.5404800176620483,0.5329792022705078,0.5486272096633911,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:10,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,512,0.5509439706802368,0.5397760152816772,0.5650751948356628,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:10,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,1024,0.9618880152702332,0.960319995880127,0.9631871938705445,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:10,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,2048,2.9309279918670654,2.9277695178985597,2.934227275848389,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:10,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,4096,10.622336387634277,10.617459297180176,10.626553535461426,"{""batch_size"": 4, ""hidden_size"": 768, ""num_heads"": 12, ""kernel_size"": 7, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:10,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,64,0.2927039861679077,0.288729590177536,0.29821439981460574,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:15,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,128,0.2951680123806,0.2902655959129333,0.3007551968097687,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:15,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,256,0.3073119968175888,0.30331520438194276,0.3113279938697815,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:15,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,512,0.31302401423454285,0.308896005153656,0.3193280100822449,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:15,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,1024,0.3047359883785248,0.29980799555778503,0.31088000535964966,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:15,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,2048,0.6286399960517883,0.6277120113372803,0.6298239827156067,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:15,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,4096,2.274079918861389,2.2696383953094483,2.2807615756988526,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:15,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,64,7.891039848327637,7.859904193878174,7.912108898162842,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:24,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,128,14.735088348388672,14.724479675292969,14.77280044555664,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:24,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,256,31.657472610473633,31.56709747314453,31.663366317749023,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:24,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,512,63.166656494140625,63.166656494140625,63.166656494140625,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:24,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,1024,126.43017578125,126.43017578125,126.43017578125,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:24,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,2048,252.40432739257812,252.40432739257812,252.40432739257812,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:24,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,4096,501.9139404296875,501.9139404296875,501.9139404296875,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:24,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,64,1.2528959512710571,1.2323967933654785,1.2715328454971313,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:31,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,128,1.2391039729118347,1.2172479629516602,1.2862399816513062,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:31,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,256,1.2568000555038452,1.233785605430603,1.2933119773864747,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:31,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,512,1.3446719646453857,1.285599946975708,1.4957120418548584,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:31,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,1024,1.295680046081543,1.2643327713012695,1.3450048208236693,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:31,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,2048,1.7712639570236206,1.7696640491485596,1.774227213859558,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:31,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,4096,6.146207809448242,6.141721630096436,6.158944129943848,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:31,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,64,8.7423677444458,8.720800399780273,8.7706880569458,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:41,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,128,16.668703079223633,16.63415107727051,16.79740104675293,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:41,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,256,32.5115852355957,32.447322082519534,32.5610237121582,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:41,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,512,64.76483154296875,64.76483154296875,64.76483154296875,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:41,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,1024,128.6598358154297,128.6598358154297,128.6598358154297,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:41,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,2048,255.95904541015625,255.95904541015625,255.95904541015625,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:41,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,4096,515.8973388671875,515.8973388671875,515.8973388671875,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:41,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,64,0.6644799709320068,0.6530431866645813,0.6826240181922913,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:42,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,128,0.666592001914978,0.6546623826026916,0.6783487796783447,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:42,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,256,0.669920027256012,0.6548351764678956,0.6889215946197509,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:42,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,512,0.6975039839744568,0.6836288094520568,0.7113920211791993,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:42,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,1024,0.699616014957428,0.6872192144393922,0.7121151924133301,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:42,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,2048,1.1469759941101074,1.1452735900878905,1.149177646636963,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:42,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,4096,3.880863904953003,3.8781951904296874,3.8849407672882084,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:42,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,64,0.5487040281295776,0.5412224054336547,0.5557375907897949,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:46,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,128,0.552592009305954,0.5447999835014343,0.5629119873046875,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:46,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,256,0.555103987455368,0.54546560049057,0.5744511961936951,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:46,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,512,0.580672025680542,0.5698879957199097,0.6001600027084351,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:46,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,1024,0.6986879706382751,0.6974080204963684,0.6997823953628539,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:46,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,2048,2.00816011428833,2.005337619781494,2.0103232383728025,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:46,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,4096,7.130911827087402,7.1263679504394535,7.134297466278077,"{""batch_size"": 2, ""hidden_size"": 1024, ""num_heads"": 16, ""kernel_size"": 9, ""dilation"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:46,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,64,0.3076159954071045,0.30279679894447326,0.3114496052265167,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:48,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,128,0.3047199994325638,0.300927996635437,0.3099200129508972,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:48,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,256,0.3068479895591736,0.30245760679244993,0.3113279938697815,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:48,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,512,0.31513598561286926,0.30997760891914367,0.3229247987270355,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:48,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,1024,0.31198400259017944,0.30742400884628296,0.3183743953704834,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:48,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,2048,0.33262400329113007,0.3312639892101288,0.3335552036762237,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:48,0.6.4
+fused_neighborhood_attention,liger,forward,speed,ms,seq_len,sequence length,4096,1.1023039817810059,1.1005760431289673,1.1036991834640504,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:48,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,64,6.187104225158691,6.177305507659912,6.194451141357422,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:56,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,128,12.469056129455566,12.451827049255371,12.514386940002442,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:56,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,256,24.972000122070312,24.941241836547853,25.024396896362305,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:56,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,512,50.10639953613281,50.10639953613281,50.10639953613281,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:56,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,1024,100.6994857788086,100.6994857788086,100.6994857788086,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:56,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,2048,200.60604858398438,200.60604858398438,200.60604858398438,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:56,0.6.4
+fused_neighborhood_attention,torch,forward,speed,ms,seq_len,sequence length,4096,402.10406494140625,402.10406494140625,402.10406494140625,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:56,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,64,1.2380159497261047,1.2136000394821167,1.2925440073013306,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:57,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,128,1.237023949623108,1.216435194015503,1.2614079713821411,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:57,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,256,1.2258880138397217,1.2073152303695678,1.2502399444580077,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:57,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,512,1.2899199724197388,1.2567039728164673,1.3232320547103882,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:57,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,1024,1.278656005859375,1.257702374458313,1.3014335870742797,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:57,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,2048,1.29203200340271,1.2728831768035889,1.329318380355835,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:57,0.6.4
+fused_neighborhood_attention,liger,full,speed,ms,seq_len,sequence length,4096,2.9953598976135254,2.991392087936401,3.000396728515625,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:42:57,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,64,7.04038405418396,7.003776168823242,7.1043647766113285,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:43:04,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,128,12.551103591918945,12.457945442199708,13.840992164611817,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:43:04,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,256,25.85785675048828,25.844550704956056,25.881241607666013,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:43:04,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,512,51.15584182739258,51.15584182739258,51.15584182739258,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:43:04,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,1024,101.17855834960938,101.17855834960938,101.17855834960938,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:43:04,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,2048,202.6748504638672,202.6748504638672,202.6748504638672,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:43:04,0.6.4
+fused_neighborhood_attention,torch,full,speed,ms,seq_len,sequence length,4096,381.1939697265625,381.1939697265625,381.1939697265625,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:43:04,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,64,0.6705919802188873,0.6556288242340088,0.6995264053344726,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:43:05,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,128,0.6734079718589783,0.6589183807373047,0.7026624202728271,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:43:05,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,256,0.6601600050926208,0.6503104209899903,0.681542420387268,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:43:05,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,512,0.7053120136260986,0.6907200217247009,0.7330880165100098,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:43:05,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,1024,0.716079980134964,0.6988352179527283,0.7466367840766907,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:43:05,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,2048,0.7027999758720398,0.6904831886291505,0.7152959823608398,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:43:05,0.6.4
+fused_neighborhood_attention,liger,backward,speed,ms,seq_len,sequence length,4096,1.8866080045700073,1.8840576171875,1.8894335746765136,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:43:05,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,64,0.5522559881210327,0.5435392141342164,0.563648009300232,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:43:08,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,128,0.5486720204353333,0.5404544115066529,0.5595263957977294,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:43:08,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,256,0.5558080077171326,0.5445759892463684,0.5874239802360535,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:43:08,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,512,0.5719040036201477,0.5639103889465332,0.5846912026405334,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:43:08,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,1024,0.5828480124473572,0.5741055846214295,0.5937152147293091,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:43:08,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,2048,1.0823359489440918,1.0813696146011353,1.0844095945358276,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:43:08,0.6.4
+fused_neighborhood_attention,torch,backward,speed,ms,seq_len,sequence length,4096,3.6391040086746216,3.6375679969787598,3.642751932144165,"{""batch_size"": 2, ""hidden_size"": 512, ""num_heads"": 8, ""kernel_size"": 7, ""dilation"": 2, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:43:08,0.6.4
+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 HBM3,2025-12-19 00:43:08,0.6.4
+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 HBM3,2025-12-19 00:43:08,0.6.4
+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 HBM3,2025-12-19 00:43:08,0.6.4
+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 HBM3,2025-12-19 00:43:08,0.6.4
+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 HBM3,2025-12-19 00:43:08,0.6.4
+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 HBM3,2025-12-19 00:43:08,0.6.4
+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 HBM3,2025-12-19 00:43:08,0.6.4
+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 HBM3,2025-12-19 00:43:16,0.6.4
+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 HBM3,2025-12-19 00:43:16,0.6.4
+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 HBM3,2025-12-19 00:43:16,0.6.4
+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 HBM3,2025-12-19 00:43:16,0.6.4
+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 HBM3,2025-12-19 00:43:16,0.6.4
+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 HBM3,2025-12-19 00:43:16,0.6.4
+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 HBM3,2025-12-19 00:43:16,0.6.4
+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 HBM3,2025-12-19 00:43:17,0.6.4
+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 HBM3,2025-12-19 00:43:17,0.6.4
+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 HBM3,2025-12-19 00:43:17,0.6.4
+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 HBM3,2025-12-19 00:43:17,0.6.4
+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 HBM3,2025-12-19 00:43:17,0.6.4
+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 HBM3,2025-12-19 00:43:17,0.6.4
+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 HBM3,2025-12-19 00:43:17,0.6.4
+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 HBM3,2025-12-19 00:43:25,0.6.4
+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 HBM3,2025-12-19 00:43:25,0.6.4
+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 HBM3,2025-12-19 00:43:25,0.6.4
+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 HBM3,2025-12-19 00:43:25,0.6.4
+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 HBM3,2025-12-19 00:43:25,0.6.4
+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 HBM3,2025-12-19 00:43:25,0.6.4
+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 HBM3,2025-12-19 00:43:25,0.6.4
+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 HBM3,2025-12-19 00:43:26,0.6.4
+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 HBM3,2025-12-19 00:43:26,0.6.4
+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 HBM3,2025-12-19 00:43:26,0.6.4
+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 HBM3,2025-12-19 00:43:26,0.6.4
+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 HBM3,2025-12-19 00:43:26,0.6.4
+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 HBM3,2025-12-19 00:43:26,0.6.4
+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 HBM3,2025-12-19 00:43:26,0.6.4
+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 HBM3,2025-12-19 00:43:37,0.6.4
+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 HBM3,2025-12-19 00:43:37,0.6.4
+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 HBM3,2025-12-19 00:43:37,0.6.4
+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 HBM3,2025-12-19 00:43:37,0.6.4
+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 HBM3,2025-12-19 00:43:37,0.6.4
+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 HBM3,2025-12-19 00:43:37,0.6.4
+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 HBM3,2025-12-19 00:43:37,0.6.4
+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 HBM3,2025-12-19 00:43:37,0.6.4
+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 HBM3,2025-12-19 00:43:37,0.6.4
+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 HBM3,2025-12-19 00:43:37,0.6.4
+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 HBM3,2025-12-19 00:43:37,0.6.4
+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 HBM3,2025-12-19 00:43:37,0.6.4
+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 HBM3,2025-12-19 00:43:37,0.6.4
+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 HBM3,2025-12-19 00:43:37,0.6.4
+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 HBM3,2025-12-19 00:43:45,0.6.4
+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 HBM3,2025-12-19 00:43:45,0.6.4
+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 HBM3,2025-12-19 00:43:45,0.6.4
+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 HBM3,2025-12-19 00:43:45,0.6.4
+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 HBM3,2025-12-19 00:43:45,0.6.4
+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 HBM3,2025-12-19 00:43:45,0.6.4
+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 HBM3,2025-12-19 00:43:45,0.6.4
+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 HBM3,2025-12-19 00:43:45,0.6.4
+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 HBM3,2025-12-19 00:43:45,0.6.4
+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 HBM3,2025-12-19 00:43:45,0.6.4
+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 HBM3,2025-12-19 00:43:45,0.6.4
+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 HBM3,2025-12-19 00:43:45,0.6.4
+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 HBM3,2025-12-19 00:43:45,0.6.4
+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 HBM3,2025-12-19 00:43:45,0.6.4
+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 HBM3,2025-12-19 00:43:53,0.6.4
+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 HBM3,2025-12-19 00:43:53,0.6.4
+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 HBM3,2025-12-19 00:43:53,0.6.4
+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 HBM3,2025-12-19 00:43:53,0.6.4
+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 HBM3,2025-12-19 00:43:53,0.6.4
+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 HBM3,2025-12-19 00:43:53,0.6.4
+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 HBM3,2025-12-19 00:43:53,0.6.4
+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 HBM3,2025-12-19 00:43:54,0.6.4
+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 HBM3,2025-12-19 00:43:54,0.6.4
+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 HBM3,2025-12-19 00:43:54,0.6.4
+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 HBM3,2025-12-19 00:43:54,0.6.4
+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 HBM3,2025-12-19 00:43:54,0.6.4
+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 HBM3,2025-12-19 00:43:54,0.6.4
+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 HBM3,2025-12-19 00:43:54,0.6.4
+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 HBM3,2025-12-19 00:44:02,0.6.4
+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 HBM3,2025-12-19 00:44:02,0.6.4
+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 HBM3,2025-12-19 00:44:02,0.6.4
+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 HBM3,2025-12-19 00:44:02,0.6.4
+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 HBM3,2025-12-19 00:44:02,0.6.4
+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 HBM3,2025-12-19 00:44:02,0.6.4
+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 HBM3,2025-12-19 00:44:02,0.6.4
+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 HBM3,2025-12-19 00:44:02,0.6.4
+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 HBM3,2025-12-19 00:44:02,0.6.4
+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 HBM3,2025-12-19 00:44:02,0.6.4
+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 HBM3,2025-12-19 00:44:02,0.6.4
+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 HBM3,2025-12-19 00:44:02,0.6.4
+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 HBM3,2025-12-19 00:44:02,0.6.4
+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 HBM3,2025-12-19 00:44:02,0.6.4
+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 HBM3,2025-12-19 00:44:13,0.6.4
+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 HBM3,2025-12-19 00:44:13,0.6.4
+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 HBM3,2025-12-19 00:44:13,0.6.4
+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 HBM3,2025-12-19 00:44:13,0.6.4
+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 HBM3,2025-12-19 00:44:13,0.6.4
+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 HBM3,2025-12-19 00:44:13,0.6.4
+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 HBM3,2025-12-19 00:44:13,0.6.4
+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 HBM3,2025-12-19 00:44:13,0.6.4
+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 HBM3,2025-12-19 00:44:13,0.6.4
+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 HBM3,2025-12-19 00:44:13,0.6.4
+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 HBM3,2025-12-19 00:44:13,0.6.4
+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 HBM3,2025-12-19 00:44:13,0.6.4
+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 HBM3,2025-12-19 00:44:13,0.6.4
+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 HBM3,2025-12-19 00:44:13,0.6.4
+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 HBM3,2025-12-19 00:44:21,0.6.4
+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 HBM3,2025-12-19 00:44:21,0.6.4
+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 HBM3,2025-12-19 00:44:21,0.6.4
+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 HBM3,2025-12-19 00:44:21,0.6.4
+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 HBM3,2025-12-19 00:44:21,0.6.4
+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 HBM3,2025-12-19 00:44:21,0.6.4
+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 HBM3,2025-12-19 00:44:21,0.6.4
+geglu,liger,full,speed,ms,T,sequence length,1024,9.674976348876953,9.674976348876953,9.674976348876953,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:32,0.6.4
+geglu,liger,full,speed,ms,T,sequence length,2048,21.090112686157227,21.090112686157227,21.090112686157227,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:32,0.6.4
+geglu,liger,full,speed,ms,T,sequence length,4096,39.65193557739258,39.65193557739258,39.65193557739258,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:32,0.6.4
+geglu,liger,full,speed,ms,T,sequence length,8192,79.71398162841797,79.71398162841797,79.71398162841797,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:32,0.6.4
+geglu,huggingface,full,speed,ms,T,sequence length,1024,10.810912132263184,10.810912132263184,10.810912132263184,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:36,0.6.4
+geglu,huggingface,full,speed,ms,T,sequence length,2048,21.573503494262695,21.573503494262695,21.573503494262695,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:36,0.6.4
+geglu,huggingface,full,speed,ms,T,sequence length,4096,40.58147048950195,40.58147048950195,40.58147048950195,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:36,0.6.4
+geglu,huggingface,full,speed,ms,T,sequence length,8192,80.83721923828125,80.83721923828125,80.83721923828125,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:36,0.6.4
+geglu,liger,forward,speed,ms,T,sequence length,1024,3.1767361164093018,3.175603294372559,3.176851272583008,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:39,0.6.4
+geglu,liger,forward,speed,ms,T,sequence length,2048,6.614175796508789,6.614175796508789,6.614175796508789,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:39,0.6.4
+geglu,liger,forward,speed,ms,T,sequence length,4096,13.699872016906738,13.699872016906738,13.699872016906738,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:39,0.6.4
+geglu,liger,forward,speed,ms,T,sequence length,8192,27.388959884643555,27.388959884643555,27.388959884643555,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:39,0.6.4
+geglu,huggingface,forward,speed,ms,T,sequence length,1024,3.348031997680664,3.199788808822632,3.3888703823089603,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:42,0.6.4
+geglu,huggingface,forward,speed,ms,T,sequence length,2048,6.84006404876709,6.84006404876709,6.84006404876709,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:42,0.6.4
+geglu,huggingface,forward,speed,ms,T,sequence length,4096,13.612031936645508,13.612031936645508,13.612031936645508,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:42,0.6.4
+geglu,huggingface,forward,speed,ms,T,sequence length,8192,27.772031784057617,27.772031784057617,27.772031784057617,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:42,0.6.4
+geglu,liger,backward,speed,ms,T,sequence length,1024,6.2454400062561035,6.2454400062561035,6.2454400062561035,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:46,0.6.4
+geglu,liger,backward,speed,ms,T,sequence length,2048,13.793503761291504,13.793503761291504,13.793503761291504,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:46,0.6.4
+geglu,liger,backward,speed,ms,T,sequence length,4096,27.594688415527344,27.594688415527344,27.594688415527344,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:46,0.6.4
+geglu,liger,backward,speed,ms,T,sequence length,8192,51.9299201965332,51.9299201965332,51.9299201965332,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:46,0.6.4
+geglu,huggingface,backward,speed,ms,T,sequence length,1024,6.946112155914307,6.946112155914307,6.946112155914307,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:49,0.6.4
+geglu,huggingface,backward,speed,ms,T,sequence length,2048,13.842047691345215,13.842047691345215,13.842047691345215,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:49,0.6.4
+geglu,huggingface,backward,speed,ms,T,sequence length,4096,27.07859230041504,27.07859230041504,27.07859230041504,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:49,0.6.4
+geglu,huggingface,backward,speed,ms,T,sequence length,8192,54.397216796875,54.397216796875,54.397216796875,"{""bsz"": 8, ""hidden_size"": 4096, ""intermediate_size"": 11008, ""hidden_act"": ""gelu_pytorch_tanh"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:44:49,0.6.4
+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 HBM3,2025-12-19 00:44:52,0.6.4
+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 HBM3,2025-12-19 00:44:52,0.6.4
+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 HBM3,2025-12-19 00:44:52,0.6.4
+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 HBM3,2025-12-19 00:44:52,0.6.4
+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 HBM3,2025-12-19 00:44:54,0.6.4
+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 HBM3,2025-12-19 00:44:54,0.6.4
+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 HBM3,2025-12-19 00:44:54,0.6.4
+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 HBM3,2025-12-19 00:44:54,0.6.4
+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 HBM3,2025-12-19 00:44:57,0.6.4
+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 HBM3,2025-12-19 00:44:57,0.6.4
+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 HBM3,2025-12-19 00:44:57,0.6.4
+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 HBM3,2025-12-19 00:44:57,0.6.4
+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 HBM3,2025-12-19 00:45:00,0.6.4
+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 HBM3,2025-12-19 00:45:00,0.6.4
+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 HBM3,2025-12-19 00:45:00,0.6.4
+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 HBM3,2025-12-19 00:45:00,0.6.4
+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 HBM3,2025-12-19 00:45:02,0.6.4
+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 HBM3,2025-12-19 00:45:02,0.6.4
+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 HBM3,2025-12-19 00:45:02,0.6.4
+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 HBM3,2025-12-19 00:45:02,0.6.4
+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 HBM3,2025-12-19 00:45:05,0.6.4
+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 HBM3,2025-12-19 00:45:05,0.6.4
+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 HBM3,2025-12-19 00:45:05,0.6.4
+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 HBM3,2025-12-19 00:45:05,0.6.4
+group_norm,liger,forward,speed,ms,C,num_channels,32,0.021727999672293663,0.02115200087428093,0.02252800017595291,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:13,0.6.4
+group_norm,liger,forward,speed,ms,C,num_channels,64,0.030047999694943428,0.029823999851942062,0.030368000268936157,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:13,0.6.4
+group_norm,liger,forward,speed,ms,C,num_channels,128,0.04508800059556961,0.044704001396894455,0.04560000076889992,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:13,0.6.4
+group_norm,liger,forward,speed,ms,C,num_channels,256,0.07315199822187424,0.07251200079917908,0.0737600028514862,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:13,0.6.4
+group_norm,liger,forward,speed,ms,C,num_channels,512,0.13523200154304504,0.133760005235672,0.1367039978504181,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:13,0.6.4
+group_norm,liger,forward,speed,ms,C,num_channels,1024,0.2603839933872223,0.2573759853839874,0.26307201385498047,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:13,0.6.4
+group_norm,liger,forward,speed,ms,C,num_channels,2048,0.5226240158081055,0.5206080079078674,0.5243840217590332,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:13,0.6.4
+group_norm,huggingface,forward,speed,ms,C,num_channels,32,0.028575999662280083,0.028351999819278717,0.028831999748945236,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:17,0.6.4
+group_norm,huggingface,forward,speed,ms,C,num_channels,64,0.04355200007557869,0.043296001851558685,0.04387199878692627,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:17,0.6.4
+group_norm,huggingface,forward,speed,ms,C,num_channels,128,0.07913599908351898,0.07865600287914276,0.07948800176382065,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:17,0.6.4
+group_norm,huggingface,forward,speed,ms,C,num_channels,256,0.14112000167369843,0.14070400595664978,0.14188799262046814,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:17,0.6.4
+group_norm,huggingface,forward,speed,ms,C,num_channels,512,0.2627840042114258,0.26233598589897156,0.263264000415802,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:17,0.6.4
+group_norm,huggingface,forward,speed,ms,C,num_channels,1024,0.5097600221633911,0.5093759894371033,0.5101439952850342,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:17,0.6.4
+group_norm,huggingface,forward,speed,ms,C,num_channels,2048,1.0018559694290161,1.0013760328292847,1.0023679733276367,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:17,0.6.4
+group_norm,liger,full,speed,ms,C,num_channels,32,0.21542400121688843,0.20698240399360657,0.22755840420722961,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:20,0.6.4
+group_norm,liger,full,speed,ms,C,num_channels,64,0.21326400339603424,0.205062398314476,0.2247679978609085,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:20,0.6.4
+group_norm,liger,full,speed,ms,C,num_channels,128,0.21000000089406967,0.20259200036525726,0.21882879734039307,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:20,0.6.4
+group_norm,liger,full,speed,ms,C,num_channels,256,0.2110079973936081,0.2040511965751648,0.22032000124454498,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:20,0.6.4
+group_norm,liger,full,speed,ms,C,num_channels,512,0.36711999773979187,0.36528640389442446,0.36902400851249695,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:20,0.6.4
+group_norm,liger,full,speed,ms,C,num_channels,1024,0.7041280269622803,0.7010175943374634,0.7071743845939635,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:20,0.6.4
+group_norm,liger,full,speed,ms,C,num_channels,2048,1.4182080030441284,1.4157119989395142,1.4205440282821655,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:20,0.6.4
+group_norm,huggingface,full,speed,ms,C,num_channels,32,0.10067199915647507,0.09356799721717834,0.10796800255775452,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:24,0.6.4
+group_norm,huggingface,full,speed,ms,C,num_channels,64,0.14883199334144592,0.14771200716495514,0.14975999295711517,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:24,0.6.4
+group_norm,huggingface,full,speed,ms,C,num_channels,128,0.22838400304317474,0.22700800001621246,0.2292799949645996,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:24,0.6.4
+group_norm,huggingface,full,speed,ms,C,num_channels,256,0.376800000667572,0.3755840063095093,0.3776960074901581,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:24,0.6.4
+group_norm,huggingface,full,speed,ms,C,num_channels,512,0.6676480174064636,0.6662399768829346,0.6685760021209717,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:24,0.6.4
+group_norm,huggingface,full,speed,ms,C,num_channels,1024,1.2510240077972412,1.2499264001846313,1.252128005027771,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:24,0.6.4
+group_norm,huggingface,full,speed,ms,C,num_channels,2048,2.411007881164551,2.4101696491241453,2.41212797164917,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:24,0.6.4
+group_norm,liger,backward,speed,ms,C,num_channels,32,0.09859199821949005,0.09043200314044952,0.10655999928712845,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:27,0.6.4
+group_norm,liger,backward,speed,ms,C,num_channels,64,0.09763199836015701,0.0902400016784668,0.10488960146903993,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:27,0.6.4
+group_norm,liger,backward,speed,ms,C,num_channels,128,0.09984000027179718,0.09408000111579895,0.1058880016207695,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:27,0.6.4
+group_norm,liger,backward,speed,ms,C,num_channels,256,0.1311040073633194,0.1295360028743744,0.1324159950017929,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:27,0.6.4
+group_norm,liger,backward,speed,ms,C,num_channels,512,0.23552000522613525,0.23481599986553192,0.23625600337982178,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:27,0.6.4
+group_norm,liger,backward,speed,ms,C,num_channels,1024,0.4506880044937134,0.44924798607826233,0.4522239863872528,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:27,0.6.4
+group_norm,liger,backward,speed,ms,C,num_channels,2048,0.9005759954452515,0.8997439742088318,0.9015872001647949,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:27,0.6.4
+group_norm,huggingface,backward,speed,ms,C,num_channels,32,0.0652799978852272,0.062111999839544296,0.07248000055551529,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:30,0.6.4
+group_norm,huggingface,backward,speed,ms,C,num_channels,64,0.1103999987244606,0.10982400178909302,0.110944002866745,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:30,0.6.4
+group_norm,huggingface,backward,speed,ms,C,num_channels,128,0.1518400013446808,0.15123200416564941,0.15244799852371216,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:30,0.6.4
+group_norm,huggingface,backward,speed,ms,C,num_channels,256,0.23999999463558197,0.23923200368881226,0.2407039999961853,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:30,0.6.4
+group_norm,huggingface,backward,speed,ms,C,num_channels,512,0.4081279933452606,0.40745601058006287,0.40883201360702515,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:30,0.6.4
+group_norm,huggingface,backward,speed,ms,C,num_channels,1024,0.7453759908676147,0.744704008102417,0.7460479736328125,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:30,0.6.4
+group_norm,huggingface,backward,speed,ms,C,num_channels,2048,1.412608027458191,1.411903977394104,1.413260817527771,"{""M"": 128, ""H"": 512, ""channels_per_group"": 4, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:30,0.6.4
+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 HBM3,2025-12-19 00:45:31,0.6.4
+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 HBM3,2025-12-19 00:45:31,0.6.4
+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 HBM3,2025-12-19 00:45:31,0.6.4
+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 HBM3,2025-12-19 00:45:31,0.6.4
+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 HBM3,2025-12-19 00:45:31,0.6.4
+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 HBM3,2025-12-19 00:45:31,0.6.4
+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 HBM3,2025-12-19 00:45:31,0.6.4
+jsd,liger,full,memory,MB,V,vocab size,4096,768.00341796875,768.00341796875,768.00341796875,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:02,0.6.4
+jsd,liger,full,memory,MB,V,vocab size,8192,1536.00341796875,1536.00341796875,1536.00341796875,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:02,0.6.4
+jsd,liger,full,memory,MB,V,vocab size,16384,3072.0048828125,3072.0048828125,3072.0048828125,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:02,0.6.4
+jsd,liger,full,memory,MB,V,vocab size,32768,6144.0087890625,6144.0087890625,6144.0087890625,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:02,0.6.4
+jsd,liger,full,memory,MB,V,vocab size,65536,12288.0166015625,12288.0166015625,12288.0166015625,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:02,0.6.4
+jsd,liger,full,memory,MB,V,vocab size,131072,24576.015625,24576.015625,24576.015625,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:02,0.6.4
+jsd,torch,full,memory,MB,V,vocab size,4096,1664.0009765625,1664.0009765625,1664.0009765625,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:03,0.6.4
+jsd,torch,full,memory,MB,V,vocab size,8192,3328.0009765625,3328.0009765625,3328.0009765625,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:03,0.6.4
+jsd,torch,full,memory,MB,V,vocab size,16384,6656.0009765625,6656.0009765625,6656.0009765625,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:03,0.6.4
+jsd,torch,full,memory,MB,V,vocab size,32768,13312.0009765625,13312.0009765625,13312.0009765625,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:03,0.6.4
+jsd,torch,full,memory,MB,V,vocab size,65536,26624.0,26624.0,26624.0,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:03,0.6.4
+jsd,torch,full,memory,MB,V,vocab size,131072,53248.0,53248.0,53248.0,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:03,0.6.4
+jsd,liger,forward,speed,ms,V,vocab size,4096,0.3006719946861267,0.3001280128955841,0.3011839985847473,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:05,0.6.4
+jsd,liger,forward,speed,ms,V,vocab size,8192,0.5946559906005859,0.5924351930618286,0.5963071823120117,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:05,0.6.4
+jsd,liger,forward,speed,ms,V,vocab size,16384,8.495648384094238,8.493215560913086,8.499391555786133,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:05,0.6.4
+jsd,liger,forward,speed,ms,V,vocab size,32768,17.423072814941406,17.41607093811035,17.426995086669923,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:05,0.6.4
+jsd,liger,forward,speed,ms,V,vocab size,65536,35.42673683166504,35.42236862182617,35.431105041503905,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:05,0.6.4
+jsd,liger,forward,speed,ms,V,vocab size,131072,71.0533447265625,71.0533447265625,71.0533447265625,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:05,0.6.4
+jsd,torch,forward,speed,ms,V,vocab size,4096,1.3618559837341309,1.3603520393371582,1.362720012664795,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:07,0.6.4
+jsd,torch,forward,speed,ms,V,vocab size,8192,2.6401760578155518,2.638751983642578,2.641184091567993,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:07,0.6.4
+jsd,torch,forward,speed,ms,V,vocab size,16384,5.18836784362793,5.186284732818604,5.189683055877685,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:07,0.6.4
+jsd,torch,forward,speed,ms,V,vocab size,32768,10.289119720458984,10.285426902770997,10.29006061553955,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:07,0.6.4
+jsd,torch,forward,speed,ms,V,vocab size,65536,20.480591773986816,20.476844787597656,20.48509407043457,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:07,0.6.4
+jsd,torch,forward,speed,ms,V,vocab size,131072,40.95091247558594,40.95075912475586,40.95106582641601,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:07,0.6.4
+jsd,liger,backward,speed,ms,V,vocab size,4096,0.32328000664711,0.32124799489974976,0.3258495986461639,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:08,0.6.4
+jsd,liger,backward,speed,ms,V,vocab size,8192,0.5878719985485077,0.5863999724388123,0.5910847902297973,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:08,0.6.4
+jsd,liger,backward,speed,ms,V,vocab size,16384,1.2113920450210571,1.2094719409942627,1.2159871816635133,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:08,0.6.4
+jsd,liger,backward,speed,ms,V,vocab size,32768,2.5391359329223633,2.537657594680786,2.5415424346923827,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:08,0.6.4
+jsd,liger,backward,speed,ms,V,vocab size,65536,5.060671806335449,5.049638557434083,5.0831808090209964,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:08,0.6.4
+jsd,liger,backward,speed,ms,V,vocab size,131072,10.047103881835938,10.042374420166016,10.049279785156251,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:08,0.6.4
+jsd,torch,backward,speed,ms,V,vocab size,4096,1.856160044670105,1.854591965675354,1.8566720485687256,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:09,0.6.4
+jsd,torch,backward,speed,ms,V,vocab size,8192,3.6360479593276978,3.6345279216766357,3.6374080181121826,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:09,0.6.4
+jsd,torch,backward,speed,ms,V,vocab size,16384,7.285791873931885,7.282617568969727,7.287219142913818,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:09,0.6.4
+jsd,torch,backward,speed,ms,V,vocab size,32768,14.664512157440186,14.662272453308105,14.665056228637695,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:09,0.6.4
+jsd,torch,backward,speed,ms,V,vocab size,65536,29.249984741210938,29.249274063110352,29.252038955688477,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:09,0.6.4
+jsd,torch,backward,speed,ms,V,vocab size,131072,58.48591995239258,58.48591995239258,58.48591995239258,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:09,0.6.4
+jsd,liger,full,speed,ms,V,vocab size,4096,0.6212480068206787,0.6197119951248169,0.623744010925293,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:11,0.6.4
+jsd,liger,full,speed,ms,V,vocab size,8192,1.178272008895874,1.1755712032318115,1.1816320419311523,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:11,0.6.4
+jsd,liger,full,speed,ms,V,vocab size,16384,9.717248439788818,9.708985328674316,9.721299362182618,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:11,0.6.4
+jsd,liger,full,speed,ms,V,vocab size,32768,19.99608039855957,19.958572769165038,20.048525619506837,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:11,0.6.4
+jsd,liger,full,speed,ms,V,vocab size,65536,40.511152267456055,40.49666519165039,40.525639343261716,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:11,0.6.4
+jsd,liger,full,speed,ms,V,vocab size,131072,81.17308807373047,81.17308807373047,81.17308807373047,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:11,0.6.4
+jsd,torch,full,speed,ms,V,vocab size,4096,3.2110079526901245,3.2095360279083254,3.21213436126709,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:13,0.6.4
+jsd,torch,full,speed,ms,V,vocab size,8192,6.263552188873291,6.262444686889648,6.266028881072998,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:13,0.6.4
+jsd,torch,full,speed,ms,V,vocab size,16384,12.464192390441895,12.461190032958983,12.467168045043945,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:13,0.6.4
+jsd,torch,full,speed,ms,V,vocab size,32768,24.937503814697266,24.934028244018556,24.9412483215332,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:13,0.6.4
+jsd,torch,full,speed,ms,V,vocab size,65536,49.73119926452637,49.72711029052735,49.73528823852539,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:13,0.6.4
+jsd,torch,full,speed,ms,V,vocab size,131072,99.43929290771484,99.43929290771484,99.43929290771484,"{""B"": 4, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:13,0.6.4
+kl_div,liger,full,memory,MB,V,vocab size,4096,384.0009765625,384.0009765625,384.0009765625,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:20,0.6.4
+kl_div,liger,full,memory,MB,V,vocab size,8192,768.0009765625,768.0009765625,768.0009765625,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:20,0.6.4
+kl_div,liger,full,memory,MB,V,vocab size,16384,1536.0009765625,1536.0009765625,1536.0009765625,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:20,0.6.4
+kl_div,liger,full,memory,MB,V,vocab size,32768,3072.0009765625,3072.0009765625,3072.0009765625,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:20,0.6.4
+kl_div,liger,full,memory,MB,V,vocab size,65536,6144.0009765625,6144.0009765625,6144.0009765625,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:20,0.6.4
+kl_div,liger,full,memory,MB,V,vocab size,131072,12288.0009765625,12288.0009765625,12288.0009765625,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:20,0.6.4
+kl_div,torch,full,memory,MB,V,vocab size,4096,448.0,448.0,448.0,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:20,0.6.4
+kl_div,torch,full,memory,MB,V,vocab size,8192,896.0,896.0,896.0,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:20,0.6.4
+kl_div,torch,full,memory,MB,V,vocab size,16384,1792.0,1792.0,1792.0,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:20,0.6.4
+kl_div,torch,full,memory,MB,V,vocab size,32768,3584.0,3584.0,3584.0,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:20,0.6.4
+kl_div,torch,full,memory,MB,V,vocab size,65536,7168.0,7168.0,7168.0,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:20,0.6.4
+kl_div,torch,full,memory,MB,V,vocab size,131072,14336.0,14336.0,14336.0,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:20,0.6.4
+kl_div,liger,forward,speed,ms,V,vocab size,4096,0.06924799829721451,0.0681023970246315,0.07062400132417679,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:21,0.6.4
+kl_div,liger,forward,speed,ms,V,vocab size,8192,0.11113599687814713,0.11015039682388306,0.11272960156202316,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:21,0.6.4
+kl_div,liger,forward,speed,ms,V,vocab size,16384,0.19707199931144714,0.19620479941368102,0.19871999323368073,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:21,0.6.4
+kl_div,liger,forward,speed,ms,V,vocab size,32768,0.368831992149353,0.3675200045108795,0.3703039884567261,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:21,0.6.4
+kl_div,liger,forward,speed,ms,V,vocab size,65536,0.7117440104484558,0.710207998752594,0.7134720087051392,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:21,0.6.4
+kl_div,liger,forward,speed,ms,V,vocab size,131072,1.4027520418167114,1.3996992111206055,1.4064640283584593,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:21,0.6.4
+kl_div,torch,forward,speed,ms,V,vocab size,4096,0.23057600110769272,0.22980480194091796,0.23117439448833466,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:21,0.6.4
+kl_div,torch,forward,speed,ms,V,vocab size,8192,0.43116798996925354,0.4305535972118378,0.43197440505027773,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:21,0.6.4
+kl_div,torch,forward,speed,ms,V,vocab size,16384,0.8260320127010345,0.8251775860786438,0.8268863797187804,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:21,0.6.4
+kl_div,torch,forward,speed,ms,V,vocab size,32768,1.6149120330810547,1.6136768102645873,1.6160255670547485,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:21,0.6.4
+kl_div,torch,forward,speed,ms,V,vocab size,65536,3.192528009414673,3.191641616821289,3.1946239948272708,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:21,0.6.4
+kl_div,torch,forward,speed,ms,V,vocab size,131072,6.3448638916015625,6.342745494842529,6.3480318069458015,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:21,0.6.4
+kl_div,liger,backward,speed,ms,V,vocab size,4096,0.2884800136089325,0.28542719483375545,0.29095680117607114,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:22,0.6.4
+kl_div,liger,backward,speed,ms,V,vocab size,8192,0.5090240240097046,0.5057088017463685,0.5129664182662964,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:22,0.6.4
+kl_div,liger,backward,speed,ms,V,vocab size,16384,1.0013760328292847,1.0003520250320435,1.0034879446029663,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:22,0.6.4
+kl_div,liger,backward,speed,ms,V,vocab size,32768,2.0259199142456055,2.023884725570679,2.0283647537231446,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:22,0.6.4
+kl_div,liger,backward,speed,ms,V,vocab size,65536,4.021072149276733,4.020249748229981,4.0223743438720705,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:22,0.6.4
+kl_div,liger,backward,speed,ms,V,vocab size,131072,7.993824005126953,7.990464115142823,7.998937702178956,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:22,0.6.4
+kl_div,torch,backward,speed,ms,V,vocab size,4096,0.2606399953365326,0.2598080039024353,0.2611840069293976,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:23,0.6.4
+kl_div,torch,backward,speed,ms,V,vocab size,8192,0.497408002614975,0.49647998809814453,0.49801599979400635,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:23,0.6.4
+kl_div,torch,backward,speed,ms,V,vocab size,16384,1.0190880298614502,1.0181440114974976,1.020134401321411,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:23,0.6.4
+kl_div,torch,backward,speed,ms,V,vocab size,32768,2.1056320667266846,2.104300832748413,2.1069760799407957,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:23,0.6.4
+kl_div,torch,backward,speed,ms,V,vocab size,65536,4.211328029632568,4.209350490570069,4.212576198577881,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:23,0.6.4
+kl_div,torch,backward,speed,ms,V,vocab size,131072,8.404767990112305,8.400351524353027,8.409024238586426,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:23,0.6.4
+kl_div,liger,full,speed,ms,V,vocab size,4096,0.35337600111961365,0.3513343870639801,0.36778879165649414,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:24,0.6.4
+kl_div,liger,full,speed,ms,V,vocab size,8192,0.6183200180530548,0.615014398097992,0.622822380065918,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:24,0.6.4
+kl_div,liger,full,speed,ms,V,vocab size,16384,1.1939200162887573,1.1900864362716674,1.1981055974960326,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:24,0.6.4
+kl_div,liger,full,speed,ms,V,vocab size,32768,2.388192057609558,2.3855552673339844,2.3940288066864013,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:24,0.6.4
+kl_div,liger,full,speed,ms,V,vocab size,65536,4.727791786193848,4.72649621963501,4.729529476165771,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:24,0.6.4
+kl_div,liger,full,speed,ms,V,vocab size,131072,9.391951560974121,9.384281349182128,9.39541130065918,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:24,0.6.4
+kl_div,torch,full,speed,ms,V,vocab size,4096,0.4861280024051666,0.4854080080986023,0.4870719909667969,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:25,0.6.4
+kl_div,torch,full,speed,ms,V,vocab size,8192,0.9254720211029053,0.9243711948394775,0.9265407800674439,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:25,0.6.4
+kl_div,torch,full,speed,ms,V,vocab size,16384,1.8390719890594482,1.8380160331726074,1.8401600122451782,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:25,0.6.4
+kl_div,torch,full,speed,ms,V,vocab size,32768,3.716096043586731,3.7151999473571777,3.7178239822387695,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:25,0.6.4
+kl_div,torch,full,speed,ms,V,vocab size,65536,7.4040961265563965,7.402079963684082,7.405619144439697,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:25,0.6.4
+kl_div,torch,full,speed,ms,V,vocab size,131072,14.750352382659912,14.748703956604004,14.752351760864258,"{""B"": 8, ""T"": 512}",NVIDIA H100 80GB HBM3,2025-12-19 00:47:25,0.6.4
+kto_loss,liger,forward,speed,ms,B,Batch Size (B),2,4.019231796264648,4.016672134399414,4.021920204162598,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:48:00,0.6.4
+kto_loss,liger,forward,speed,ms,B,Batch Size (B),4,7.9488959312438965,7.933504104614258,7.954880237579346,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:48:00,0.6.4
+kto_loss,liger,forward,speed,ms,B,Batch Size (B),8,15.792799949645996,15.754752159118652,15.821855545043945,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:48:00,0.6.4
+kto_loss,liger,forward,speed,ms,B,Batch Size (B),16,31.41974449157715,31.41826591491699,31.466399765014646,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:48:00,0.6.4
+kto_loss,liger,forward,speed,ms,B,Batch Size (B),32,62.87785720825195,62.87785720825195,62.87785720825195,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:48:00,0.6.4
+kto_loss,huggingface,forward,speed,ms,B,Batch Size (B),2,3.6755200624465942,3.66975998878479,3.6813440322875977,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:48:16,0.6.4
+kto_loss,huggingface,forward,speed,ms,B,Batch Size (B),4,6.782047986984253,6.770003032684326,6.786912155151367,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:48:16,0.6.4
+kto_loss,huggingface,forward,speed,ms,B,Batch Size (B),8,12.643296241760254,12.640447807312011,12.676454544067383,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:48:16,0.6.4
+kto_loss,huggingface,forward,speed,ms,B,Batch Size (B),16,24.529888153076172,24.525132751464845,24.538099670410155,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:48:16,0.6.4
+kto_loss,huggingface,forward,speed,ms,B,Batch Size (B),32,48.52615928649902,48.51693420410157,48.53538436889649,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:48:16,0.6.4
+kto_loss,liger,backward,speed,ms,B,Batch Size (B),2,0.3102239966392517,0.30902400612831116,0.3116991937160492,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:48:32,0.6.4
+kto_loss,liger,backward,speed,ms,B,Batch Size (B),4,0.3102879971265793,0.3086720108985901,0.31331199407577515,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:48:32,0.6.4
+kto_loss,liger,backward,speed,ms,B,Batch Size (B),8,0.30847999453544617,0.30742400884628296,0.3097600042819977,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:48:32,0.6.4
+kto_loss,liger,backward,speed,ms,B,Batch Size (B),16,0.3087199926376343,0.30774399638175964,0.30971519947052,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:48:32,0.6.4
+kto_loss,liger,backward,speed,ms,B,Batch Size (B),32,0.3097279965877533,0.30873599648475647,0.3109567940235138,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:48:32,0.6.4
+kto_loss,huggingface,backward,speed,ms,B,Batch Size (B),2,2.095871925354004,2.0921151638031006,2.100729560852051,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:48:47,0.6.4
+kto_loss,huggingface,backward,speed,ms,B,Batch Size (B),4,3.610960006713867,3.6081600189208984,3.6196160316467285,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:48:47,0.6.4
+kto_loss,huggingface,backward,speed,ms,B,Batch Size (B),8,6.6907360553741455,6.681913661956787,6.704076957702637,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:48:47,0.6.4
+kto_loss,huggingface,backward,speed,ms,B,Batch Size (B),16,12.78934383392334,12.780109214782716,12.803212547302246,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:48:47,0.6.4
+kto_loss,huggingface,backward,speed,ms,B,Batch Size (B),32,24.968095779418945,24.968058013916014,24.974546813964842,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:48:47,0.6.4
+kto_loss,liger,full,speed,ms,B,Batch Size (B),2,6.062047958374023,6.0478081703186035,6.0726399421691895,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:49:03,0.6.4
+kto_loss,liger,full,speed,ms,B,Batch Size (B),4,10.011967658996582,9.98172836303711,10.018163108825684,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:49:03,0.6.4
+kto_loss,liger,full,speed,ms,B,Batch Size (B),8,18.31065559387207,18.23449592590332,18.343999481201173,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:49:03,0.6.4
+kto_loss,liger,full,speed,ms,B,Batch Size (B),16,33.80751991271973,33.80533065795899,33.80970916748047,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:49:03,0.6.4
+kto_loss,liger,full,speed,ms,B,Batch Size (B),32,65.41535949707031,65.41535949707031,65.41535949707031,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:49:03,0.6.4
+kto_loss,huggingface,full,speed,ms,B,Batch Size (B),2,6.168767929077148,6.161888122558594,6.2039361000061035,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:49:18,0.6.4
+kto_loss,huggingface,full,speed,ms,B,Batch Size (B),4,10.511103630065918,10.495129585266113,10.52010898590088,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:49:18,0.6.4
+kto_loss,huggingface,full,speed,ms,B,Batch Size (B),8,19.385343551635742,19.379936599731444,19.421561431884765,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:49:18,0.6.4
+kto_loss,huggingface,full,speed,ms,B,Batch Size (B),16,37.33406448364258,37.32775650024414,37.340372467041014,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:49:18,0.6.4
+kto_loss,huggingface,full,speed,ms,B,Batch Size (B),32,73.53727722167969,73.53727722167969,73.53727722167969,"{""T"": 512, ""H"": 1024, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16"", ""bias"": true, ""beta"": 0.1, ""ignore_index"": 42}",NVIDIA H100 80GB HBM3,2025-12-19 00:49:18,0.6.4
+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 HBM3,2025-12-19 00:49:38,0.6.4
+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 HBM3,2025-12-19 00:49:38,0.6.4
+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 HBM3,2025-12-19 00:49:38,0.6.4
+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 HBM3,2025-12-19 00:49:38,0.6.4
+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 HBM3,2025-12-19 00:49:38,0.6.4
+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 HBM3,2025-12-19 00:49:54,0.6.4
+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 HBM3,2025-12-19 00:49:54,0.6.4
+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 HBM3,2025-12-19 00:49:54,0.6.4
+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 HBM3,2025-12-19 00:49:54,0.6.4
+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 HBM3,2025-12-19 00:49:54,0.6.4
+layer_norm,liger,forward,speed,ms,N,hidden size,1024,0.018271999433636665,0.018015999346971512,0.018719999119639397,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:03,0.6.4
+layer_norm,liger,forward,speed,ms,N,hidden size,2048,0.029440000653266907,0.029055999591946602,0.03014400042593479,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:03,0.6.4
+layer_norm,liger,forward,speed,ms,N,hidden size,4096,0.05152000114321709,0.05081599950790405,0.05225599929690361,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:03,0.6.4
+layer_norm,liger,forward,speed,ms,N,hidden size,8192,0.09641599655151367,0.09604480117559433,0.09678719937801361,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:03,0.6.4
+layer_norm,liger,forward,speed,ms,N,hidden size,16384,0.18627199530601501,0.1855040043592453,0.18678399920463562,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:03,0.6.4
+layer_norm,huggingface,forward,speed,ms,N,hidden size,1024,0.02316799946129322,0.022911999374628067,0.023455999791622162,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:06,0.6.4
+layer_norm,huggingface,forward,speed,ms,N,hidden size,2048,0.036768000572919846,0.03654399886727333,0.03705599904060364,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:06,0.6.4
+layer_norm,huggingface,forward,speed,ms,N,hidden size,4096,0.06905599683523178,0.06838399916887283,0.06963200122117996,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:06,0.6.4
+layer_norm,huggingface,forward,speed,ms,N,hidden size,8192,0.1496960073709488,0.14908799529075623,0.15036800503730774,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:06,0.6.4
+layer_norm,huggingface,forward,speed,ms,N,hidden size,16384,0.29683199524879456,0.2961919903755188,0.2975040078163147,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:06,0.6.4
+layer_norm,liger,backward,speed,ms,N,hidden size,1024,0.15358399599790573,0.13210879564285277,0.30158080458641057,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:09,0.6.4
+layer_norm,liger,backward,speed,ms,N,hidden size,2048,0.1311040073633194,0.1236800029873848,0.14032000303268433,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:09,0.6.4
+layer_norm,liger,backward,speed,ms,N,hidden size,4096,0.132192000746727,0.12556800246238708,0.13837439715862274,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:09,0.6.4
+layer_norm,liger,backward,speed,ms,N,hidden size,8192,0.17027199268341064,0.16924799978733063,0.17139199376106262,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:09,0.6.4
+layer_norm,liger,backward,speed,ms,N,hidden size,16384,0.6066880226135254,0.6034560203552246,0.6120255947113037,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:09,0.6.4
+layer_norm,huggingface,backward,speed,ms,N,hidden size,1024,0.06019200012087822,0.05923200026154518,0.06108799949288368,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:12,0.6.4
+layer_norm,huggingface,backward,speed,ms,N,hidden size,2048,0.10041599720716476,0.09916800260543823,0.10182400047779083,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:12,0.6.4
+layer_norm,huggingface,backward,speed,ms,N,hidden size,4096,0.18111999332904816,0.1799360066652298,0.18214400112628937,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:12,0.6.4
+layer_norm,huggingface,backward,speed,ms,N,hidden size,8192,0.34217599034309387,0.34092798829078674,0.34329599142074585,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:12,0.6.4
+layer_norm,huggingface,backward,speed,ms,N,hidden size,16384,0.660176008939743,0.6589120030403137,0.6613759994506836,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:12,0.6.4
+layer_norm,liger,full,speed,ms,N,hidden size,1024,0.22511999309062958,0.21767680048942567,0.23432320058345796,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:14,0.6.4
+layer_norm,liger,full,speed,ms,N,hidden size,2048,0.22870400547981262,0.22073599696159363,0.23900799453258514,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:14,0.6.4
+layer_norm,liger,full,speed,ms,N,hidden size,4096,0.22572800517082214,0.21893759965896606,0.23510399460792542,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:14,0.6.4
+layer_norm,liger,full,speed,ms,N,hidden size,8192,0.26396799087524414,0.26265600323677063,0.26524800062179565,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:14,0.6.4
+layer_norm,liger,full,speed,ms,N,hidden size,16384,0.7919199764728546,0.7879679799079895,0.7969215869903564,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:14,0.6.4
+layer_norm,huggingface,full,speed,ms,N,hidden size,1024,0.07923199981451035,0.0783040001988411,0.08022399991750717,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:16,0.6.4
+layer_norm,huggingface,full,speed,ms,N,hidden size,2048,0.13247999548912048,0.13116799294948578,0.13395200669765472,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:16,0.6.4
+layer_norm,huggingface,full,speed,ms,N,hidden size,4096,0.24614399671554565,0.24512000381946564,0.24716800451278687,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:16,0.6.4
+layer_norm,huggingface,full,speed,ms,N,hidden size,8192,0.48796799778938293,0.4868031978607178,0.48924800753593445,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:16,0.6.4
+layer_norm,huggingface,full,speed,ms,N,hidden size,16384,0.9530879855155945,0.9517760276794434,0.9545919895172119,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:16,0.6.4
+layer_norm,liger,full,memory,MB,N,hidden size,1024,81.09375,81.09375,81.09375,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:16,0.6.4
+layer_norm,liger,full,memory,MB,N,hidden size,2048,162.15625,162.15625,162.15625,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:16,0.6.4
+layer_norm,liger,full,memory,MB,N,hidden size,4096,324.28125,324.28125,324.28125,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:16,0.6.4
+layer_norm,liger,full,memory,MB,N,hidden size,8192,648.53125,648.53125,648.53125,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:16,0.6.4
+layer_norm,liger,full,memory,MB,N,hidden size,16384,1297.03125,1297.03125,1297.03125,"{""M"": 4096, ""dtype"": ""torch.float32"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:16,0.6.4
+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 HBM3,2025-12-19 00:50:16,0.6.4
+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 HBM3,2025-12-19 00:50:16,0.6.4
+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 HBM3,2025-12-19 00:50:16,0.6.4
+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 HBM3,2025-12-19 00:50:16,0.6.4
+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 HBM3,2025-12-19 00:50:16,0.6.4
+llama4_rope,liger,forward,speed,ms,H,hidden size,512,0.062144000083208084,0.06188800185918808,0.06239999830722809,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:25,0.6.4
+llama4_rope,liger,forward,speed,ms,H,hidden size,2048,0.061535999178886414,0.060736000537872314,0.06195199862122536,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:25,0.6.4
+llama4_rope,liger,forward,speed,ms,H,hidden size,8192,0.06854400038719177,0.06812799721956253,0.0689919963479042,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:25,0.6.4
+llama4_rope,huggingface,forward,speed,ms,H,hidden size,512,0.03129599988460541,0.030432000756263733,0.03249920159578325,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:26,0.6.4
+llama4_rope,huggingface,forward,speed,ms,H,hidden size,2048,0.062144000083208084,0.0615679994225502,0.0629120022058487,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:26,0.6.4
+llama4_rope,huggingface,forward,speed,ms,H,hidden size,8192,0.20793600380420685,0.2072640061378479,0.2086080014705658,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:26,0.6.4
+llama4_rope,liger,backward,speed,ms,H,hidden size,512,0.12995200604200363,0.12256640195846558,0.13880319893360138,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:27,0.6.4
+llama4_rope,liger,backward,speed,ms,H,hidden size,2048,0.12942399829626083,0.12265600264072418,0.13705599308013916,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:27,0.6.4
+llama4_rope,liger,backward,speed,ms,H,hidden size,8192,0.13120000064373016,0.12265600264072418,0.14419199526309967,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:27,0.6.4
+llama4_rope,huggingface,backward,speed,ms,H,hidden size,512,0.14659200608730316,0.13756799697875977,0.1598847985267639,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:28,0.6.4
+llama4_rope,huggingface,backward,speed,ms,H,hidden size,2048,0.14182400703430176,0.13411200046539307,0.14963840544223786,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:28,0.6.4
+llama4_rope,huggingface,backward,speed,ms,H,hidden size,8192,0.21276800334453583,0.21135999262332916,0.21404799818992615,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:28,0.6.4
+llama4_rope,liger,full,speed,ms,H,hidden size,512,0.2714720070362091,0.26236799359321594,0.2831679880619049,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:30,0.6.4
+llama4_rope,liger,full,speed,ms,H,hidden size,2048,0.2682560086250305,0.25925120115280154,0.27881601452827454,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:30,0.6.4
+llama4_rope,liger,full,speed,ms,H,hidden size,8192,0.26710399985313416,0.2586367905139923,0.27765120267868043,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:30,0.6.4
+llama4_rope,huggingface,full,speed,ms,H,hidden size,512,0.25484800338745117,0.247161602973938,0.2652864038944244,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:31,0.6.4
+llama4_rope,huggingface,full,speed,ms,H,hidden size,2048,0.2566719949245453,0.24920959770679474,0.26760320067405696,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:31,0.6.4
+llama4_rope,huggingface,full,speed,ms,H,hidden size,8192,0.4172160029411316,0.41625601053237915,0.4182400107383728,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:31,0.6.4
+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 HBM3,2025-12-19 00:50:31,0.6.4
+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 HBM3,2025-12-19 00:50:31,0.6.4
+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 HBM3,2025-12-19 00:50:31,0.6.4
+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 HBM3,2025-12-19 00:50:31,0.6.4
+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 HBM3,2025-12-19 00:50:31,0.6.4
+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 HBM3,2025-12-19 00:50:31,0.6.4
+llama4_rope,liger,forward,speed,ms,T,sequence length,1024,0.04473600164055824,0.04412800073623657,0.04601600021123886,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:33,0.6.4
+llama4_rope,liger,forward,speed,ms,T,sequence length,2048,0.0681919977068901,0.06755200028419495,0.06883200258016586,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:33,0.6.4
+llama4_rope,liger,forward,speed,ms,T,sequence length,4096,0.11641599982976913,0.11606399714946747,0.1167680025100708,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:33,0.6.4
+llama4_rope,liger,forward,speed,ms,T,sequence length,8192,0.21686400473117828,0.21619200706481934,0.21756799519062042,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:33,0.6.4
+llama4_rope,liger,forward,speed,ms,T,sequence length,16384,0.4180159866809845,0.41737601161003113,0.41862401366233826,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:33,0.6.4
+llama4_rope,huggingface,forward,speed,ms,T,sequence length,1024,0.11766400188207626,0.1170239970088005,0.11846400052309036,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:35,0.6.4
+llama4_rope,huggingface,forward,speed,ms,T,sequence length,2048,0.20790399610996246,0.20732800662517548,0.2085759937763214,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:35,0.6.4
+llama4_rope,huggingface,forward,speed,ms,T,sequence length,4096,0.388480007648468,0.38758400082588196,0.38921600580215454,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:35,0.6.4
+llama4_rope,huggingface,forward,speed,ms,T,sequence length,8192,0.7454400062561035,0.7446720004081726,0.7464896202087403,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:35,0.6.4
+llama4_rope,huggingface,forward,speed,ms,T,sequence length,16384,1.4626399874687195,1.4617919921875,1.4637759923934937,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:35,0.6.4
+llama4_rope,liger,backward,speed,ms,T,sequence length,1024,0.13179200142621994,0.12569600343704224,0.13896319568157198,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:38,0.6.4
+llama4_rope,liger,backward,speed,ms,T,sequence length,2048,0.1305920034646988,0.12377600371837616,0.1383103966712952,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:38,0.6.4
+llama4_rope,liger,backward,speed,ms,T,sequence length,4096,0.13504000008106232,0.12845439612865447,0.14313599467277527,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:38,0.6.4
+llama4_rope,liger,backward,speed,ms,T,sequence length,8192,0.21667200326919556,0.2160000056028366,0.21743999421596527,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:38,0.6.4
+llama4_rope,liger,backward,speed,ms,T,sequence length,16384,0.4182719886302948,0.4177280068397522,0.4188480079174042,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:38,0.6.4
+llama4_rope,huggingface,backward,speed,ms,T,sequence length,1024,0.1440960019826889,0.1364479959011078,0.15219199657440186,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:40,0.6.4
+llama4_rope,huggingface,backward,speed,ms,T,sequence length,2048,0.21212799847126007,0.21084800362586975,0.21347199380397797,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:40,0.6.4
+llama4_rope,huggingface,backward,speed,ms,T,sequence length,4096,0.40031999349594116,0.39907199144363403,0.40137600898742676,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:40,0.6.4
+llama4_rope,huggingface,backward,speed,ms,T,sequence length,8192,0.7590399980545044,0.7576959729194641,0.7602751970291137,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:40,0.6.4
+llama4_rope,huggingface,backward,speed,ms,T,sequence length,16384,1.4923839569091797,1.4909696340560914,1.4934719800949097,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:40,0.6.4
+llama4_rope,liger,full,speed,ms,T,sequence length,1024,0.270687997341156,0.2611455976963043,0.2876288115978241,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:42,0.6.4
+llama4_rope,liger,full,speed,ms,T,sequence length,2048,0.2736479938030243,0.26342400908470154,0.29087998867034914,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:42,0.6.4
+llama4_rope,liger,full,speed,ms,T,sequence length,4096,0.273824006319046,0.26492801308631897,0.2834751963615418,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:42,0.6.4
+llama4_rope,liger,full,speed,ms,T,sequence length,8192,0.4312640130519867,0.43001601099967957,0.43284480571746825,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:42,0.6.4
+llama4_rope,liger,full,speed,ms,T,sequence length,16384,0.8343679904937744,0.8335360288619995,0.8359424114227295,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:42,0.6.4
+llama4_rope,huggingface,full,speed,ms,T,sequence length,1024,0.2518079876899719,0.24497279822826384,0.26005120277404786,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:44,0.6.4
+llama4_rope,huggingface,full,speed,ms,T,sequence length,2048,0.41702398657798767,0.4160960018634796,0.4180287897586823,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:44,0.6.4
+llama4_rope,huggingface,full,speed,ms,T,sequence length,4096,0.7854400277137756,0.7843520045280457,0.7865279912948608,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:44,0.6.4
+llama4_rope,huggingface,full,speed,ms,T,sequence length,8192,1.5017759799957275,1.500864028930664,1.5028159618377686,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:44,0.6.4
+llama4_rope,huggingface,full,speed,ms,T,sequence length,16384,2.9523839950561523,2.951200008392334,2.95414400100708,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:50:44,0.6.4
+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 HBM3,2025-12-19 00:50:44,0.6.4
+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 HBM3,2025-12-19 00:50:44,0.6.4
+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 HBM3,2025-12-19 00:50:44,0.6.4
+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 HBM3,2025-12-19 00:50:44,0.6.4
+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 HBM3,2025-12-19 00:50:44,0.6.4
+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 HBM3,2025-12-19 00:50:44,0.6.4
+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 HBM3,2025-12-19 00:50:44,0.6.4
+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 HBM3,2025-12-19 00:50:44,0.6.4
+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 HBM3,2025-12-19 00:50:44,0.6.4
+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 HBM3,2025-12-19 00:50:44,0.6.4
+multi_token_attention,liger,forward,speed,ms,L,sequence length,32,0.12060800194740295,0.11612799763679504,0.12522879987955093,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:37,0.6.4
+multi_token_attention,liger,forward,speed,ms,L,sequence length,64,0.11560000106692314,0.11347199976444244,0.1189119964838028,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:37,0.6.4
+multi_token_attention,liger,forward,speed,ms,L,sequence length,128,0.11868800222873688,0.11690880060195923,0.12124799937009811,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:37,0.6.4
+multi_token_attention,liger,forward,speed,ms,L,sequence length,256,0.15902400016784668,0.15849600732326508,0.15964800119400024,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:37,0.6.4
+multi_token_attention,liger,forward,speed,ms,L,sequence length,512,0.6983360052108765,0.6966911792755126,0.6999743819236754,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:37,0.6.4
+multi_token_attention,torch,forward,speed,ms,L,sequence length,32,0.15360000729560852,0.1517760008573532,0.1573760062456131,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:38,0.6.4
+multi_token_attention,torch,forward,speed,ms,L,sequence length,64,0.14692799746990204,0.1460864007472992,0.14908799529075623,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:38,0.6.4
+multi_token_attention,torch,forward,speed,ms,L,sequence length,128,0.15248000621795654,0.15158399939537048,0.15398399531841278,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:38,0.6.4
+multi_token_attention,torch,forward,speed,ms,L,sequence length,256,0.15065599977970123,0.14985600113868713,0.15238399803638458,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:38,0.6.4
+multi_token_attention,torch,forward,speed,ms,L,sequence length,512,0.19643200188875198,0.19473279714584352,0.19804799556732178,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:38,0.6.4
+multi_token_attention,liger,full,speed,ms,L,sequence length,32,0.6088799834251404,0.5960319876670838,0.6261376023292542,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:47,0.6.4
+multi_token_attention,liger,full,speed,ms,L,sequence length,64,0.6614719927310944,0.6489983916282654,0.682700800895691,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:47,0.6.4
+multi_token_attention,liger,full,speed,ms,L,sequence length,128,0.6658239960670471,0.6519680142402648,0.6918272137641908,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:47,0.6.4
+multi_token_attention,liger,full,speed,ms,L,sequence length,256,0.6671839952468872,0.6462015986442566,0.6888703942298889,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:47,0.6.4
+multi_token_attention,liger,full,speed,ms,L,sequence length,512,1.3234559893608093,1.3198016166687012,1.3268544197082521,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:47,0.6.4
+multi_token_attention,torch,full,speed,ms,L,sequence length,32,0.4135200083255768,0.40388480424880985,0.4336319983005524,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:47,0.6.4
+multi_token_attention,torch,full,speed,ms,L,sequence length,64,0.44307199120521545,0.4336639940738678,0.469760000705719,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:47,0.6.4
+multi_token_attention,torch,full,speed,ms,L,sequence length,128,0.44281598925590515,0.43317119479179383,0.46184959411621096,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:47,0.6.4
+multi_token_attention,torch,full,speed,ms,L,sequence length,256,0.4477279931306839,0.4370175898075104,0.4662079930305481,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:47,0.6.4
+multi_token_attention,torch,full,speed,ms,L,sequence length,512,0.6339040100574493,0.6275392055511475,0.6440512180328368,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:47,0.6.4
+multi_token_attention,liger,backward,speed,ms,L,sequence length,32,0.3051519989967346,0.29836800694465637,0.31439998745918274,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:48,0.6.4
+multi_token_attention,liger,backward,speed,ms,L,sequence length,64,0.34064000844955444,0.3294528126716614,0.3533375859260559,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:48,0.6.4
+multi_token_attention,liger,backward,speed,ms,L,sequence length,128,0.33820798993110657,0.32894080877304077,0.3521983981132507,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:48,0.6.4
+multi_token_attention,liger,backward,speed,ms,L,sequence length,256,0.3420160114765167,0.33129599690437317,0.35671040415763855,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:48,0.6.4
+multi_token_attention,liger,backward,speed,ms,L,sequence length,512,0.6308479905128479,0.6290496230125427,0.6331520080566406,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:48,0.6.4
+multi_token_attention,torch,backward,speed,ms,L,sequence length,32,0.1329760029911995,0.1258944034576416,0.139615997672081,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:49,0.6.4
+multi_token_attention,torch,backward,speed,ms,L,sequence length,64,0.15324799716472626,0.14536959826946258,0.1609216034412384,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:49,0.6.4
+multi_token_attention,torch,backward,speed,ms,L,sequence length,128,0.15329600125551224,0.14491519331932068,0.16050559282302856,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:49,0.6.4
+multi_token_attention,torch,backward,speed,ms,L,sequence length,256,0.1642560064792633,0.16328320205211638,0.16527999937534332,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:49,0.6.4
+multi_token_attention,torch,backward,speed,ms,L,sequence length,512,0.34992000460624695,0.3483327984809875,0.3514431953430176,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:51:49,0.6.4
+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 HBM3,2025-12-19 00:51:49,0.6.4
+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 HBM3,2025-12-19 00:51:49,0.6.4
+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 HBM3,2025-12-19 00:51:49,0.6.4
+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 HBM3,2025-12-19 00:51:49,0.6.4
+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 HBM3,2025-12-19 00:51:49,0.6.4
+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 HBM3,2025-12-19 00:51:49,0.6.4
+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 HBM3,2025-12-19 00:51:49,0.6.4
+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 HBM3,2025-12-19 00:51:49,0.6.4
+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 HBM3,2025-12-19 00:51:49,0.6.4
+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 HBM3,2025-12-19 00:51:49,0.6.4
+poly_norm,liger,forward,speed,ms,H,hidden size,1024,0.01017600018531084,0.009881599806249142,0.010432000271975994,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:29,0.6.4
+poly_norm,liger,forward,speed,ms,H,hidden size,2048,0.014336000196635723,0.013952000066637993,0.014751999638974667,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:29,0.6.4
+poly_norm,liger,forward,speed,ms,H,hidden size,4096,0.020416000857949257,0.020255999639630318,0.020576000213623047,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:29,0.6.4
+poly_norm,liger,forward,speed,ms,H,hidden size,8192,0.03564799949526787,0.035360001027584076,0.03606399893760681,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:29,0.6.4
+poly_norm,liger,forward,speed,ms,H,hidden size,16384,0.0724480003118515,0.07184000313282013,0.07311999797821045,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:29,0.6.4
+poly_norm,liger,forward,speed,ms,H,hidden size,32768,0.6611199975013733,0.6597440242767334,0.6625279784202576,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:29,0.6.4
+poly_norm,huggingface,forward,speed,ms,H,hidden size,1024,0.16415999829769135,0.1621440052986145,0.17161600291728973,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:32,0.6.4
+poly_norm,huggingface,forward,speed,ms,H,hidden size,2048,0.2808000147342682,0.2796480059623718,0.281823992729187,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:32,0.6.4
+poly_norm,huggingface,forward,speed,ms,H,hidden size,4096,0.573855996131897,0.572767972946167,0.574944019317627,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:32,0.6.4
+poly_norm,huggingface,forward,speed,ms,H,hidden size,8192,1.03929603099823,1.0380480289459229,1.0405759811401367,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:32,0.6.4
+poly_norm,huggingface,forward,speed,ms,H,hidden size,16384,1.9625279903411865,1.961087942123413,1.964038372039795,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:32,0.6.4
+poly_norm,huggingface,forward,speed,ms,H,hidden size,32768,3.7962400913238525,3.7948927879333496,3.797350358963013,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:32,0.6.4
+poly_norm,liger,full,speed,ms,H,hidden size,1024,0.2136479988694191,0.20405119955539702,0.22903039753437043,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:37,0.6.4
+poly_norm,liger,full,speed,ms,H,hidden size,2048,0.2131199985742569,0.2039552003145218,0.22746239900588988,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:37,0.6.4
+poly_norm,liger,full,speed,ms,H,hidden size,4096,0.2130720019340515,0.20361599326133728,0.22787199914455414,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:37,0.6.4
+poly_norm,liger,full,speed,ms,H,hidden size,8192,0.21299200505018234,0.20384640097618104,0.22952320277690888,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:37,0.6.4
+poly_norm,liger,full,speed,ms,H,hidden size,16384,0.2160319983959198,0.2070399969816208,0.22912639677524566,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:37,0.6.4
+poly_norm,liger,full,speed,ms,H,hidden size,32768,1.3603839874267578,1.3536768198013305,1.364902400970459,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:37,0.6.4
+poly_norm,huggingface,full,speed,ms,H,hidden size,1024,0.8132960200309753,0.8016191840171814,0.8399807810783386,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:40,0.6.4
+poly_norm,huggingface,full,speed,ms,H,hidden size,2048,1.0093439817428589,1.0071872234344483,1.011423945426941,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:40,0.6.4
+poly_norm,huggingface,full,speed,ms,H,hidden size,4096,1.984287977218628,1.9815552473068239,1.9862464189529419,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:40,0.6.4
+poly_norm,huggingface,full,speed,ms,H,hidden size,8192,3.639039993286133,3.636313581466675,3.6412479877471924,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:40,0.6.4
+poly_norm,huggingface,full,speed,ms,H,hidden size,16384,6.866399765014648,6.863743782043457,6.869696140289307,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:40,0.6.4
+poly_norm,huggingface,full,speed,ms,H,hidden size,32768,13.293184280395508,13.289766120910643,13.296902656555176,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:40,0.6.4
+poly_norm,liger,backward,speed,ms,H,hidden size,1024,0.11264000087976456,0.10531839728355408,0.12009599804878235,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:43,0.6.4
+poly_norm,liger,backward,speed,ms,H,hidden size,2048,0.11099200323224068,0.10361599922180176,0.11737599968910217,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:43,0.6.4
+poly_norm,liger,backward,speed,ms,H,hidden size,4096,0.11206399649381638,0.10492800176143646,0.1189119964838028,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:43,0.6.4
+poly_norm,liger,backward,speed,ms,H,hidden size,8192,0.11033599823713303,0.1035199984908104,0.11744000017642975,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:43,0.6.4
+poly_norm,liger,backward,speed,ms,H,hidden size,16384,0.12272000312805176,0.12191999703645706,0.12360960245132448,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:43,0.6.4
+poly_norm,liger,backward,speed,ms,H,hidden size,32768,0.7020319998264313,0.695360004901886,0.7070528149604798,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:43,0.6.4
+poly_norm,huggingface,backward,speed,ms,H,hidden size,1024,0.51283198595047,0.5056959986686707,0.5238847732543945,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:46,0.6.4
+poly_norm,huggingface,backward,speed,ms,H,hidden size,2048,0.7311999797821045,0.7291520237922668,0.7330560088157654,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:46,0.6.4
+poly_norm,huggingface,backward,speed,ms,H,hidden size,4096,1.414687991142273,1.412768006324768,1.4165120124816895,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:46,0.6.4
+poly_norm,huggingface,backward,speed,ms,H,hidden size,8192,2.601855993270874,2.599782419204712,2.6040639877319336,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:46,0.6.4
+poly_norm,huggingface,backward,speed,ms,H,hidden size,16384,4.906895875930786,4.9042047500610355,4.909279823303223,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:46,0.6.4
+poly_norm,huggingface,backward,speed,ms,H,hidden size,32768,9.495391845703125,9.49289608001709,9.497759819030762,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:46,0.6.4
+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 HBM3,2025-12-19 00:52:46,0.6.4
+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 HBM3,2025-12-19 00:52:46,0.6.4
+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 HBM3,2025-12-19 00:52:46,0.6.4
+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 HBM3,2025-12-19 00:52:46,0.6.4
+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 HBM3,2025-12-19 00:52:46,0.6.4
+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 HBM3,2025-12-19 00:52:46,0.6.4
+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 HBM3,2025-12-19 00:52:47,0.6.4
+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 HBM3,2025-12-19 00:52:47,0.6.4
+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 HBM3,2025-12-19 00:52:47,0.6.4
+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 HBM3,2025-12-19 00:52:47,0.6.4
+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 HBM3,2025-12-19 00:52:47,0.6.4
+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 HBM3,2025-12-19 00:52:47,0.6.4
+rms_norm,liger,forward,speed,ms,H,hidden size,1024,0.009631999768316746,0.009344000369310379,0.00979200005531311,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:58,0.6.4
+rms_norm,liger,forward,speed,ms,H,hidden size,2048,0.013024000450968742,0.012864000163972378,0.013248000293970108,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:58,0.6.4
+rms_norm,liger,forward,speed,ms,H,hidden size,4096,0.01929599978029728,0.01897599920630455,0.019731199368834496,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:58,0.6.4
+rms_norm,liger,forward,speed,ms,H,hidden size,8192,0.03155200183391571,0.031168000772595406,0.03209599852561951,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:58,0.6.4
+rms_norm,liger,forward,speed,ms,H,hidden size,16384,0.06371200084686279,0.06310400366783142,0.0644799992442131,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:58,0.6.4
+rms_norm,liger,forward,speed,ms,H,hidden size,32768,0.1098880022764206,0.10956799983978271,0.11024639755487442,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:52:58,0.6.4
+rms_norm,huggingface,forward,speed,ms,H,hidden size,1024,0.0568000003695488,0.05593600124120712,0.05782400071620941,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:01,0.6.4
+rms_norm,huggingface,forward,speed,ms,H,hidden size,2048,0.0843840017914772,0.08310399949550629,0.08582399785518646,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:01,0.6.4
+rms_norm,huggingface,forward,speed,ms,H,hidden size,4096,0.17136000096797943,0.1695680022239685,0.1728000044822693,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:01,0.6.4
+rms_norm,huggingface,forward,speed,ms,H,hidden size,8192,0.3127039968967438,0.31094399094581604,0.31407999992370605,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:01,0.6.4
+rms_norm,huggingface,forward,speed,ms,H,hidden size,16384,0.5839519798755646,0.5821120142936707,0.5854719877243042,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:01,0.6.4
+rms_norm,huggingface,forward,speed,ms,H,hidden size,32768,1.1227200031280518,1.12118399143219,1.1239807844161989,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:01,0.6.4
+rms_norm,liger,full,speed,ms,H,hidden size,1024,0.20364800095558167,0.19493119716644286,0.2155776023864746,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:06,0.6.4
+rms_norm,liger,full,speed,ms,H,hidden size,2048,0.1982399970293045,0.18982400000095367,0.2081536054611206,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:06,0.6.4
+rms_norm,liger,full,speed,ms,H,hidden size,4096,0.1971520036458969,0.1889919936656952,0.20771199464797974,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:06,0.6.4
+rms_norm,liger,full,speed,ms,H,hidden size,8192,0.19862399995326996,0.1910911977291107,0.20826880335807801,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:06,0.6.4
+rms_norm,liger,full,speed,ms,H,hidden size,16384,0.20473599433898926,0.1953023999929428,0.21691520214080812,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:06,0.6.4
+rms_norm,liger,full,speed,ms,H,hidden size,32768,0.9226399958133698,0.918995213508606,0.9262783765792847,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:06,0.6.4
+rms_norm,huggingface,full,speed,ms,H,hidden size,1024,0.2552320063114166,0.2481600046157837,0.2654847979545593,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:09,0.6.4
+rms_norm,huggingface,full,speed,ms,H,hidden size,2048,0.3134079873561859,0.31248000264167786,0.3142400085926056,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:09,0.6.4
+rms_norm,huggingface,full,speed,ms,H,hidden size,4096,0.6153280138969421,0.6143680214881897,0.6162239909172058,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:09,0.6.4
+rms_norm,huggingface,full,speed,ms,H,hidden size,8192,1.1215360164642334,1.1202944040298461,1.1227519512176514,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:09,0.6.4
+rms_norm,huggingface,full,speed,ms,H,hidden size,16384,2.1098239421844482,2.1085760593414307,2.1110079288482666,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:09,0.6.4
+rms_norm,huggingface,full,speed,ms,H,hidden size,32768,4.063184022903442,4.061888217926025,4.065216064453125,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:09,0.6.4
+rms_norm,liger,backward,speed,ms,H,hidden size,1024,0.09567999839782715,0.08864639699459076,0.10277760028839111,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:12,0.6.4
+rms_norm,liger,backward,speed,ms,H,hidden size,2048,0.09559999778866768,0.08862079679965973,0.10227199643850327,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:12,0.6.4
+rms_norm,liger,backward,speed,ms,H,hidden size,4096,0.0958079993724823,0.08854400366544724,0.10263679921627045,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:12,0.6.4
+rms_norm,liger,backward,speed,ms,H,hidden size,8192,0.09455999732017517,0.08822400122880936,0.10188800096511841,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:12,0.6.4
+rms_norm,liger,backward,speed,ms,H,hidden size,16384,0.10950399935245514,0.10767360031604767,0.11100800335407257,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:12,0.6.4
+rms_norm,liger,backward,speed,ms,H,hidden size,32768,0.8157119750976562,0.8119232177734375,0.8187455892562867,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:12,0.6.4
+rms_norm,huggingface,backward,speed,ms,H,hidden size,1024,0.16067199409008026,0.15343999862670898,0.16742399334907532,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:15,0.6.4
+rms_norm,huggingface,backward,speed,ms,H,hidden size,2048,0.22761599719524384,0.2265663981437683,0.22860799729824066,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:15,0.6.4
+rms_norm,huggingface,backward,speed,ms,H,hidden size,4096,0.44812801480293274,0.44710400700569153,0.44921600818634033,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:15,0.6.4
+rms_norm,huggingface,backward,speed,ms,H,hidden size,8192,0.8151360154151917,0.8142399787902832,0.8160319924354553,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:15,0.6.4
+rms_norm,huggingface,backward,speed,ms,H,hidden size,16384,1.5269759893417358,1.5258879661560059,1.5283775806427002,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:15,0.6.4
+rms_norm,huggingface,backward,speed,ms,H,hidden size,32768,2.9411840438842773,2.9398528575897216,2.9425791263580323,"{""M"": 2048, ""dtype"": ""torch.bfloat16"", ""eps"": 1e-06}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:15,0.6.4
+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 HBM3,2025-12-19 00:53:15,0.6.4
+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 HBM3,2025-12-19 00:53:15,0.6.4
+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 HBM3,2025-12-19 00:53:15,0.6.4
+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 HBM3,2025-12-19 00:53:15,0.6.4
+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 HBM3,2025-12-19 00:53:15,0.6.4
+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 HBM3,2025-12-19 00:53:15,0.6.4
+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 HBM3,2025-12-19 00:53:15,0.6.4
+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 HBM3,2025-12-19 00:53:15,0.6.4
+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 HBM3,2025-12-19 00:53:15,0.6.4
+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 HBM3,2025-12-19 00:53:15,0.6.4
+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 HBM3,2025-12-19 00:53:15,0.6.4
+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 HBM3,2025-12-19 00:53:15,0.6.4
+rope,liger,forward,speed,ms,H,hidden size,512,0.008191999979317188,0.007903999648988247,0.008448000065982342,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:22,0.6.4
+rope,liger,forward,speed,ms,H,hidden size,2048,0.014431999996304512,0.013824000023305416,0.01486719995737077,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:22,0.6.4
+rope,liger,forward,speed,ms,H,hidden size,8192,0.03455999866127968,0.034143999218940735,0.03526400029659271,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:22,0.6.4
+rope,huggingface,forward,speed,ms,H,hidden size,512,0.06036800146102905,0.0597120001912117,0.061216000467538834,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:24,0.6.4
+rope,huggingface,forward,speed,ms,H,hidden size,2048,0.1069440022110939,0.1064319983124733,0.107744000852108,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:24,0.6.4
+rope,huggingface,forward,speed,ms,H,hidden size,8192,0.32947200536727905,0.3285120129585266,0.3309119939804077,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:24,0.6.4
+rope,liger,backward,speed,ms,H,hidden size,512,0.07264000177383423,0.06499200314283371,0.07897599786520004,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:25,0.6.4
+rope,liger,backward,speed,ms,H,hidden size,2048,0.07176000252366066,0.06435199826955795,0.07898879796266556,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:25,0.6.4
+rope,liger,backward,speed,ms,H,hidden size,8192,0.07144000008702278,0.06394880264997482,0.07772800326347351,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:25,0.6.4
+rope,huggingface,backward,speed,ms,H,hidden size,512,0.18327999860048294,0.17606399953365326,0.19263360500335694,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:27,0.6.4
+rope,huggingface,backward,speed,ms,H,hidden size,2048,0.18475200235843658,0.1770304054021835,0.1936576008796692,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:27,0.6.4
+rope,huggingface,backward,speed,ms,H,hidden size,8192,0.392192006111145,0.39103359580039976,0.3936319947242737,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:27,0.6.4
+rope,liger,full,speed,ms,H,hidden size,512,0.1377280056476593,0.13120000064373016,0.14467200636863708,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:28,0.6.4
+rope,liger,full,speed,ms,H,hidden size,2048,0.1406400054693222,0.13407999277114868,0.14746880233287812,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:28,0.6.4
+rope,liger,full,speed,ms,H,hidden size,8192,0.13977600634098053,0.1329600065946579,0.14713600277900696,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:28,0.6.4
+rope,huggingface,full,speed,ms,H,hidden size,512,0.30131199955940247,0.29526400566101074,0.30789760351181034,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:29,0.6.4
+rope,huggingface,full,speed,ms,H,hidden size,2048,0.302623987197876,0.2956480085849762,0.31033599376678467,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:29,0.6.4
+rope,huggingface,full,speed,ms,H,hidden size,8192,0.7179200053215027,0.7158399820327759,0.719430410861969,"{""dtype"": ""torch.bfloat16"", ""seq_len"": 2048, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:29,0.6.4
+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 HBM3,2025-12-19 00:53:29,0.6.4
+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 HBM3,2025-12-19 00:53:29,0.6.4
+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 HBM3,2025-12-19 00:53:29,0.6.4
+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 HBM3,2025-12-19 00:53:29,0.6.4
+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 HBM3,2025-12-19 00:53:29,0.6.4
+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 HBM3,2025-12-19 00:53:29,0.6.4
+rope,liger,forward,speed,ms,T,sequence length,1024,0.01961600035429001,0.01929599978029728,0.019999999552965164,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:31,0.6.4
+rope,liger,forward,speed,ms,T,sequence length,2048,0.03455999866127968,0.034143999218940735,0.0352960005402565,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:31,0.6.4
+rope,liger,forward,speed,ms,T,sequence length,4096,0.06359999999403954,0.06259199976921082,0.06431999802589417,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:31,0.6.4
+rope,liger,forward,speed,ms,T,sequence length,8192,0.12064000219106674,0.12028799951076508,0.12115199863910675,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:31,0.6.4
+rope,liger,forward,speed,ms,T,sequence length,16384,0.23472000658512115,0.23423999547958374,0.2353920042514801,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:31,0.6.4
+rope,huggingface,forward,speed,ms,T,sequence length,1024,0.1815039962530136,0.18057599663734436,0.1831360012292862,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:34,0.6.4
+rope,huggingface,forward,speed,ms,T,sequence length,2048,0.32947200536727905,0.3285439908504486,0.3309311985969543,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:34,0.6.4
+rope,huggingface,forward,speed,ms,T,sequence length,4096,0.616703987121582,0.6157439947128296,0.6180672168731689,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:34,0.6.4
+rope,huggingface,forward,speed,ms,T,sequence length,8192,1.1895999908447266,1.1885440349578857,1.190783977508545,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:34,0.6.4
+rope,huggingface,forward,speed,ms,T,sequence length,16384,2.3249919414520264,2.3238720893859863,2.3262720108032227,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:34,0.6.4
+rope,liger,backward,speed,ms,T,sequence length,1024,0.07049600034952164,0.06294400244951248,0.07686399668455124,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:36,0.6.4
+rope,liger,backward,speed,ms,T,sequence length,2048,0.07091200351715088,0.06319999694824219,0.0772159993648529,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:36,0.6.4
+rope,liger,backward,speed,ms,T,sequence length,4096,0.07241600006818771,0.06601600348949432,0.07795199751853943,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:36,0.6.4
+rope,liger,backward,speed,ms,T,sequence length,8192,0.12099199742078781,0.12067200243473053,0.12140800058841705,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:36,0.6.4
+rope,liger,backward,speed,ms,T,sequence length,16384,0.23465600609779358,0.23420800268650055,0.23539840281009672,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:36,0.6.4
+rope,huggingface,backward,speed,ms,T,sequence length,1024,0.2149759978055954,0.21395200490951538,0.21635200083255768,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:38,0.6.4
+rope,huggingface,backward,speed,ms,T,sequence length,2048,0.3924799859523773,0.3912703931331634,0.3940800130367279,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:38,0.6.4
+rope,huggingface,backward,speed,ms,T,sequence length,4096,0.7301759719848633,0.7292799949645996,0.731872022151947,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:38,0.6.4
+rope,huggingface,backward,speed,ms,T,sequence length,8192,1.4048160314559937,1.4035840034484863,1.4062912225723267,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:38,0.6.4
+rope,huggingface,backward,speed,ms,T,sequence length,16384,2.7328639030456543,2.7314560413360596,2.734368085861206,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:38,0.6.4
+rope,liger,full,speed,ms,T,sequence length,1024,0.13702400028705597,0.13023999333381653,0.14467200636863708,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:40,0.6.4
+rope,liger,full,speed,ms,T,sequence length,2048,0.13571199774742126,0.12890880107879638,0.14203519821166993,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:40,0.6.4
+rope,liger,full,speed,ms,T,sequence length,4096,0.1390720009803772,0.1342719942331314,0.14471040666103363,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:40,0.6.4
+rope,liger,full,speed,ms,T,sequence length,8192,0.23852799832820892,0.23798400163650513,0.24006399512290955,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:40,0.6.4
+rope,liger,full,speed,ms,T,sequence length,16384,0.4670400023460388,0.46620801091194153,0.4678399860858917,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:40,0.6.4
+rope,huggingface,full,speed,ms,T,sequence length,1024,0.3925760090351105,0.39129599928855896,0.39395201206207275,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:42,0.6.4
+rope,huggingface,full,speed,ms,T,sequence length,2048,0.7169920206069946,0.7154240012168884,0.7186560034751892,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:42,0.6.4
+rope,huggingface,full,speed,ms,T,sequence length,4096,1.3440799713134766,1.3424256086349486,1.3451584100723266,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:42,0.6.4
+rope,huggingface,full,speed,ms,T,sequence length,8192,2.5915839672088623,2.5901183605194094,2.592992067337036,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:42,0.6.4
+rope,huggingface,full,speed,ms,T,sequence length,16384,5.055200099945068,5.053395366668702,5.056313705444335,"{""dtype"": ""torch.bfloat16"", ""hidden_size"": 8192, ""num_q_heads"": 32, ""num_kv_heads"": 8}",NVIDIA H100 80GB HBM3,2025-12-19 00:53:42,0.6.4
+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 HBM3,2025-12-19 00:53:42,0.6.4
+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 HBM3,2025-12-19 00:53:42,0.6.4
+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 HBM3,2025-12-19 00:53:42,0.6.4
+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 HBM3,2025-12-19 00:53:42,0.6.4
+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 HBM3,2025-12-19 00:53:42,0.6.4
+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 HBM3,2025-12-19 00:53:42,0.6.4
+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 HBM3,2025-12-19 00:53:42,0.6.4
+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 HBM3,2025-12-19 00:53:42,0.6.4
+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 HBM3,2025-12-19 00:53:42,0.6.4
+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 HBM3,2025-12-19 00:53:42,0.6.4
+fused_linear_simpo_loss,liger,forward,speed,ms,B,B,2,12.117471694946289,12.05429801940918,12.19781093597412,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:54:24,0.6.4
+fused_linear_simpo_loss,liger,forward,speed,ms,B,B,4,24.458383560180664,24.347078323364258,24.55581398010254,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:54:24,0.6.4
+fused_linear_simpo_loss,liger,forward,speed,ms,B,B,8,48.84598350524902,48.725100708007815,48.96686630249023,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:54:24,0.6.4
+fused_linear_simpo_loss,liger,forward,speed,ms,B,B,16,96.4610595703125,96.4610595703125,96.4610595703125,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:54:24,0.6.4
+fused_linear_simpo_loss,huggingface,forward,speed,ms,B,B,2,6.153088092803955,6.145535945892334,6.159526538848876,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:54:48,0.6.4
+fused_linear_simpo_loss,huggingface,forward,speed,ms,B,B,4,12.017888069152832,12.00299530029297,12.021504020690918,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:54:48,0.6.4
+fused_linear_simpo_loss,huggingface,forward,speed,ms,B,B,8,23.539631843566895,23.49867515563965,23.584671783447266,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:54:48,0.6.4
+fused_linear_simpo_loss,huggingface,forward,speed,ms,B,B,16,46.94068908691406,46.86923675537109,47.012141418457034,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:54:48,0.6.4
+fused_linear_simpo_loss,liger,full,speed,ms,B,B,2,13.83011245727539,13.803852462768555,13.908960151672364,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:55:12,0.6.4
+fused_linear_simpo_loss,liger,full,speed,ms,B,B,4,25.795679092407227,25.641292572021484,25.82159996032715,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:55:12,0.6.4
+fused_linear_simpo_loss,liger,full,speed,ms,B,B,8,49.93303871154785,49.54749298095704,50.31858444213867,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:55:12,0.6.4
+fused_linear_simpo_loss,liger,full,speed,ms,B,B,16,98.49568176269531,98.49568176269531,98.49568176269531,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:55:12,0.6.4
+fused_linear_simpo_loss,huggingface,full,speed,ms,B,B,2,17.385568618774414,17.38162612915039,17.387359619140625,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:55:37,0.6.4
+fused_linear_simpo_loss,huggingface,full,speed,ms,B,B,4,33.84657669067383,33.5502914428711,34.142861938476564,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:55:37,0.6.4
+fused_linear_simpo_loss,huggingface,full,speed,ms,B,B,8,65.47792053222656,65.47792053222656,65.47792053222656,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:55:37,0.6.4
+fused_linear_simpo_loss,huggingface,full,speed,ms,B,B,16,136.46505737304688,136.46505737304688,136.46505737304688,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:55:37,0.6.4
+fused_linear_simpo_loss,liger,backward,speed,ms,B,B,2,1.0925440192222595,1.0907520055770874,1.0959359407424927,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:56:00,0.6.4
+fused_linear_simpo_loss,liger,backward,speed,ms,B,B,4,1.103775978088379,1.1019840240478516,1.1094271898269654,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:56:00,0.6.4
+fused_linear_simpo_loss,liger,backward,speed,ms,B,B,8,1.1249439716339111,1.1241215944290162,1.126860785484314,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:56:00,0.6.4
+fused_linear_simpo_loss,liger,backward,speed,ms,B,B,16,1.1731039881706238,1.1703360080718994,1.186560034751892,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:56:00,0.6.4
+fused_linear_simpo_loss,huggingface,backward,speed,ms,B,B,2,11.15998363494873,11.15849609375,11.164019012451172,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:56:24,0.6.4
+fused_linear_simpo_loss,huggingface,backward,speed,ms,B,B,4,21.252431869506836,21.196025466918947,21.35950126647949,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:56:24,0.6.4
+fused_linear_simpo_loss,huggingface,backward,speed,ms,B,B,8,41.82228660583496,41.730654907226565,41.91391830444336,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:56:24,0.6.4
+fused_linear_simpo_loss,huggingface,backward,speed,ms,B,B,16,83.9637451171875,83.9637451171875,83.9637451171875,"{""T"": 1024, ""H"": 4096, ""V"": 128256, ""mode"": ""forward"", ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:56:24,0.6.4
+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 HBM3,2025-12-19 00:56:49,0.6.4
+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 HBM3,2025-12-19 00:56:49,0.6.4
+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 HBM3,2025-12-19 00:56:49,0.6.4
+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 HBM3,2025-12-19 00:56:49,0.6.4
+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 HBM3,2025-12-19 00:57:14,0.6.4
+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 HBM3,2025-12-19 00:57:14,0.6.4
+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 HBM3,2025-12-19 00:57:14,0.6.4
+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 HBM3,2025-12-19 00:57:14,0.6.4
+softmax,liger,forward,speed,ms,N,hidden size,128,0.007519999984651804,0.007424000184983015,0.007679999805986881,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:22,0.6.4
+softmax,liger,forward,speed,ms,N,hidden size,256,0.007872000336647034,0.007584000006318092,0.008128000423312187,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:22,0.6.4
+softmax,liger,forward,speed,ms,N,hidden size,512,0.008960000239312649,0.008704000152647495,0.009184000082314014,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:22,0.6.4
+softmax,liger,forward,speed,ms,N,hidden size,1024,0.011935999616980553,0.011680000461637974,0.012191999703645706,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:22,0.6.4
+softmax,liger,forward,speed,ms,N,hidden size,2048,0.0180479995906353,0.017855999991297722,0.01833599992096424,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:22,0.6.4
+softmax,liger,forward,speed,ms,N,hidden size,4096,0.02860799990594387,0.028255999088287354,0.02924799919128418,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:22,0.6.4
+softmax,torch,forward,speed,ms,N,hidden size,128,0.007455999962985516,0.007391999941319227,0.007584000006318092,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:26,0.6.4
+softmax,torch,forward,speed,ms,N,hidden size,256,0.008224000222980976,0.00800000037997961,0.008448000065982342,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:26,0.6.4
+softmax,torch,forward,speed,ms,N,hidden size,512,0.009983999654650688,0.009855999611318111,0.01017600018531084,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:26,0.6.4
+softmax,torch,forward,speed,ms,N,hidden size,1024,0.013344000093638897,0.013088000006973743,0.014112000353634357,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:26,0.6.4
+softmax,torch,forward,speed,ms,N,hidden size,2048,0.020096000283956528,0.019840000197291374,0.02035200037062168,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:26,0.6.4
+softmax,torch,forward,speed,ms,N,hidden size,4096,0.037151999771595,0.03683200106024742,0.037567999213933945,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:26,0.6.4
+softmax,liger,full,speed,ms,N,hidden size,128,0.11219199746847153,0.10450559854507446,0.12210560142993929,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:29,0.6.4
+softmax,liger,full,speed,ms,N,hidden size,256,0.1098880022764206,0.1024319976568222,0.11763200163841248,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:29,0.6.4
+softmax,liger,full,speed,ms,N,hidden size,512,0.10841599851846695,0.10120320320129395,0.1152447983622551,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:29,0.6.4
+softmax,liger,full,speed,ms,N,hidden size,1024,0.10937599837779999,0.10220160335302353,0.11664000153541565,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:29,0.6.4
+softmax,liger,full,speed,ms,N,hidden size,2048,0.10992000252008438,0.10239999741315842,0.11764480173587799,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:29,0.6.4
+softmax,liger,full,speed,ms,N,hidden size,4096,0.10976000130176544,0.10236799716949463,0.1177792027592659,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:29,0.6.4
+softmax,torch,full,speed,ms,N,hidden size,128,0.028192000463604927,0.019999999552965164,0.038656000047922134,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:32,0.6.4
+softmax,torch,full,speed,ms,N,hidden size,256,0.021503999829292297,0.017376000061631203,0.030086399987339984,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:32,0.6.4
+softmax,torch,full,speed,ms,N,hidden size,512,0.02287999913096428,0.021215999498963356,0.030592000111937523,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:32,0.6.4
+softmax,torch,full,speed,ms,N,hidden size,1024,0.030368000268936157,0.030079999938607216,0.030688000842928886,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:32,0.6.4
+softmax,torch,full,speed,ms,N,hidden size,2048,0.05177599936723709,0.05132799968123436,0.052480001002550125,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:32,0.6.4
+softmax,torch,full,speed,ms,N,hidden size,4096,0.11219199746847153,0.11177600175142288,0.11260800063610077,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:32,0.6.4
+softmax,liger,backward,speed,ms,N,hidden size,128,0.04419200122356415,0.03827200084924698,0.05318399891257286,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:35,0.6.4
+softmax,liger,backward,speed,ms,N,hidden size,256,0.04387199878692627,0.03814399987459183,0.052639998495578766,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:35,0.6.4
+softmax,liger,backward,speed,ms,N,hidden size,512,0.04355200007557869,0.037728000432252884,0.05252480134367943,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:35,0.6.4
+softmax,liger,backward,speed,ms,N,hidden size,1024,0.04606400057673454,0.03948799893260002,0.05527039989829064,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:35,0.6.4
+softmax,liger,backward,speed,ms,N,hidden size,2048,0.04535999894142151,0.03888640105724335,0.054496001452207565,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:35,0.6.4
+softmax,liger,backward,speed,ms,N,hidden size,4096,0.04355200007557869,0.04095999896526337,0.050464000552892685,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:35,0.6.4
+softmax,torch,backward,speed,ms,N,hidden size,128,0.010688000358641148,0.010463999584317207,0.010912000201642513,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:38,0.6.4
+softmax,torch,backward,speed,ms,N,hidden size,256,0.01244799979031086,0.012128000147640705,0.013120000250637531,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:38,0.6.4
+softmax,torch,backward,speed,ms,N,hidden size,512,0.015296000055968761,0.01484800036996603,0.015968000516295433,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:38,0.6.4
+softmax,torch,backward,speed,ms,N,hidden size,1024,0.022336000576615334,0.021888000890612602,0.02304000034928322,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:38,0.6.4
+softmax,torch,backward,speed,ms,N,hidden size,2048,0.040383998304605484,0.0398080013692379,0.04108799993991852,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:38,0.6.4
+softmax,torch,backward,speed,ms,N,hidden size,4096,0.07782399654388428,0.07708799839019775,0.07836800068616867,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:38,0.6.4
+softmax,liger,forward,speed,ms,N,hidden size,128,0.007519999984651804,0.007424000184983015,0.007648000027984381,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:41,0.6.4
+softmax,liger,forward,speed,ms,N,hidden size,256,0.007455999962985516,0.00723200011998415,0.007679999805986881,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:41,0.6.4
+softmax,liger,forward,speed,ms,N,hidden size,512,0.008031999692320824,0.007872000336647034,0.008191999979317188,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:41,0.6.4
+softmax,liger,forward,speed,ms,N,hidden size,1024,0.009119999594986439,0.008991999551653862,0.009247999638319016,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:41,0.6.4
+softmax,liger,forward,speed,ms,N,hidden size,2048,0.012256000190973282,0.012128000147640705,0.01244799979031086,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:41,0.6.4
+softmax,liger,forward,speed,ms,N,hidden size,4096,0.0180479995906353,0.017855999991297722,0.01836800016462803,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:41,0.6.4
+softmax,torch,forward,speed,ms,N,hidden size,128,0.007519999984651804,0.0072639998979866505,0.007712000049650669,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:44,0.6.4
+softmax,torch,forward,speed,ms,N,hidden size,256,0.007872000336647034,0.007807999849319458,0.00800000037997961,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:44,0.6.4
+softmax,torch,forward,speed,ms,N,hidden size,512,0.009535999968647957,0.00940799992531538,0.009727999567985535,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:44,0.6.4
+softmax,torch,forward,speed,ms,N,hidden size,1024,0.012640000320971012,0.012512000277638435,0.012768000364303589,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:44,0.6.4
+softmax,torch,forward,speed,ms,N,hidden size,2048,0.018912000581622124,0.01881599985063076,0.0191040001809597,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:44,0.6.4
+softmax,torch,forward,speed,ms,N,hidden size,4096,0.033535998314619064,0.03340800106525421,0.03376000002026558,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:44,0.6.4
+softmax,liger,full,speed,ms,N,hidden size,128,0.1061440035700798,0.09923200309276581,0.1136000007390976,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:47,0.6.4
+softmax,liger,full,speed,ms,N,hidden size,256,0.10716799646615982,0.09993600100278854,0.1149887964129448,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:47,0.6.4
+softmax,liger,full,speed,ms,N,hidden size,512,0.10713600367307663,0.10009600222110748,0.11401599645614624,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:47,0.6.4
+softmax,liger,full,speed,ms,N,hidden size,1024,0.10672000050544739,0.09979519993066788,0.11392000317573547,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:47,0.6.4
+softmax,liger,full,speed,ms,N,hidden size,2048,0.10784000158309937,0.10089600086212158,0.11481600254774094,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:47,0.6.4
+softmax,liger,full,speed,ms,N,hidden size,4096,0.1082719974219799,0.10108800232410431,0.11580800265073776,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:47,0.6.4
+softmax,torch,full,speed,ms,N,hidden size,128,0.020128000527620316,0.015296000055968761,0.027424000203609467,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:50,0.6.4
+softmax,torch,full,speed,ms,N,hidden size,256,0.02006400004029274,0.01600000075995922,0.028031999245285988,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:50,0.6.4
+softmax,torch,full,speed,ms,N,hidden size,512,0.019071999937295914,0.018688000738620758,0.02300800010561943,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:50,0.6.4
+softmax,torch,full,speed,ms,N,hidden size,1024,0.025472000241279602,0.02518399991095066,0.025760000571608543,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:50,0.6.4
+softmax,torch,full,speed,ms,N,hidden size,2048,0.0344959981739521,0.0342399999499321,0.03481600061058998,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:50,0.6.4
+softmax,torch,full,speed,ms,N,hidden size,4096,0.06675200164318085,0.0663359984755516,0.06729599833488464,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:50,0.6.4
+softmax,liger,backward,speed,ms,N,hidden size,128,0.04363200068473816,0.03846399858593941,0.051872000098228455,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:53,0.6.4
+softmax,liger,backward,speed,ms,N,hidden size,256,0.04368000105023384,0.03838080167770386,0.052191998809576035,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:53,0.6.4
+softmax,liger,backward,speed,ms,N,hidden size,512,0.04390399903059006,0.0385855995118618,0.05177599936723709,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:53,0.6.4
+softmax,liger,backward,speed,ms,N,hidden size,1024,0.048608001321554184,0.04118400067090988,0.058848001062870026,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:53,0.6.4
+softmax,liger,backward,speed,ms,N,hidden size,2048,0.05161599814891815,0.041471999138593674,0.07777919918298723,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:53,0.6.4
+softmax,liger,backward,speed,ms,N,hidden size,4096,0.04755200073122978,0.038201600313186646,0.06611199676990509,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:53,0.6.4
+softmax,torch,backward,speed,ms,N,hidden size,128,0.016383999958634377,0.011071999557316303,0.02598400041460991,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,torch,backward,speed,ms,N,hidden size,256,0.013376000337302685,0.01158399973064661,0.020927999168634415,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,torch,backward,speed,ms,N,hidden size,512,0.0144640002399683,0.013088000006973743,0.02222719974815847,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,torch,backward,speed,ms,N,hidden size,1024,0.01692800037562847,0.01616000011563301,0.017696000635623932,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,torch,backward,speed,ms,N,hidden size,2048,0.02332800067961216,0.022943999618291855,0.024191999807953835,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,torch,backward,speed,ms,N,hidden size,4096,0.04073600098490715,0.04012800008058548,0.04150399938225746,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,liger,full,memory,MB,N,hidden size,128,0.00244140625,0.00244140625,0.00244140625,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,liger,full,memory,MB,N,hidden size,256,0.0048828125,0.0048828125,0.0048828125,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,liger,full,memory,MB,N,hidden size,512,0.009765625,0.009765625,0.009765625,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,liger,full,memory,MB,N,hidden size,1024,0.01953125,0.01953125,0.01953125,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,liger,full,memory,MB,N,hidden size,2048,0.0390625,0.0390625,0.0390625,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,liger,full,memory,MB,N,hidden size,4096,0.078125,0.078125,0.078125,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,torch,full,memory,MB,N,hidden size,128,0.0029296875,0.0029296875,0.0029296875,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,torch,full,memory,MB,N,hidden size,256,0.005859375,0.005859375,0.005859375,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,torch,full,memory,MB,N,hidden size,512,0.01171875,0.01171875,0.01171875,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,torch,full,memory,MB,N,hidden size,1024,0.0234375,0.0234375,0.0234375,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,torch,full,memory,MB,N,hidden size,2048,0.046875,0.046875,0.046875,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,torch,full,memory,MB,N,hidden size,4096,0.09375,0.09375,0.09375,"{""M"": 2048, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,liger,full,memory,MB,N,hidden size,128,0.00244140625,0.00244140625,0.00244140625,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,liger,full,memory,MB,N,hidden size,256,0.00244140625,0.00244140625,0.00244140625,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,liger,full,memory,MB,N,hidden size,512,0.0048828125,0.0048828125,0.0048828125,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,liger,full,memory,MB,N,hidden size,1024,0.009765625,0.009765625,0.009765625,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,liger,full,memory,MB,N,hidden size,2048,0.01953125,0.01953125,0.01953125,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,liger,full,memory,MB,N,hidden size,4096,0.0390625,0.0390625,0.0390625,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,torch,full,memory,MB,N,hidden size,128,0.0029296875,0.0029296875,0.0029296875,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,torch,full,memory,MB,N,hidden size,256,0.0029296875,0.0029296875,0.0029296875,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,torch,full,memory,MB,N,hidden size,512,0.005859375,0.005859375,0.005859375,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,torch,full,memory,MB,N,hidden size,1024,0.01171875,0.01171875,0.01171875,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,torch,full,memory,MB,N,hidden size,2048,0.0234375,0.0234375,0.0234375,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+softmax,torch,full,memory,MB,N,hidden size,4096,0.046875,0.046875,0.046875,"{""M"": 2048, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 00:57:57,0.6.4
+sparse_multi_token_attention,liger,forward,speed,ms,L,sequence length,32,0.17348799854516983,0.1679680049419403,0.1828799992799759,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:44,0.6.4
+sparse_multi_token_attention,liger,forward,speed,ms,L,sequence length,64,0.15964800119400024,0.157151997089386,0.1635199934244156,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:44,0.6.4
+sparse_multi_token_attention,liger,forward,speed,ms,L,sequence length,128,0.16526400297880173,0.16153599321842194,0.17023999989032745,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:44,0.6.4
+sparse_multi_token_attention,liger,forward,speed,ms,L,sequence length,256,0.26390400528907776,0.2630592107772827,0.26493440866470336,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:44,0.6.4
+sparse_multi_token_attention,liger,forward,speed,ms,L,sequence length,512,0.7742720246315002,0.7721023917198181,0.7764608144760132,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:44,0.6.4
+sparse_multi_token_attention,torch,forward,speed,ms,L,sequence length,32,0.43057601153850555,0.42297598719596863,0.4424639940261841,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:45,0.6.4
+sparse_multi_token_attention,torch,forward,speed,ms,L,sequence length,64,0.4065279960632324,0.4020479917526245,0.4109887957572937,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:45,0.6.4
+sparse_multi_token_attention,torch,forward,speed,ms,L,sequence length,128,0.41648000478744507,0.41168640851974486,0.42103679776191716,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:45,0.6.4
+sparse_multi_token_attention,torch,forward,speed,ms,L,sequence length,256,0.41888000071048737,0.41519999504089355,0.42446080446243284,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:45,0.6.4
+sparse_multi_token_attention,torch,forward,speed,ms,L,sequence length,512,0.4527359902858734,0.4490559995174408,0.45743998885154724,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:45,0.6.4
+sparse_multi_token_attention,liger,full,speed,ms,L,sequence length,32,0.7069759964942932,0.6887871861457825,0.7305600047111511,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:57,0.6.4
+sparse_multi_token_attention,liger,full,speed,ms,L,sequence length,64,0.6978560090065002,0.6778880000114441,0.7368255853652954,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:57,0.6.4
+sparse_multi_token_attention,liger,full,speed,ms,L,sequence length,128,0.7229120135307312,0.7018240094184875,0.7623680233955383,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:57,0.6.4
+sparse_multi_token_attention,liger,full,speed,ms,L,sequence length,256,0.7056000232696533,0.6847167849540711,0.7252543926239015,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:57,0.6.4
+sparse_multi_token_attention,liger,full,speed,ms,L,sequence length,512,1.5431360006332397,1.5355839729309082,1.5502079725265503,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:57,0.6.4
+sparse_multi_token_attention,torch,full,speed,ms,L,sequence length,32,0.9698880016803741,0.9590463876724243,0.9892607808113099,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:58,0.6.4
+sparse_multi_token_attention,torch,full,speed,ms,L,sequence length,64,0.9800319969654083,0.9675968050956726,0.9966272234916688,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:58,0.6.4
+sparse_multi_token_attention,torch,full,speed,ms,L,sequence length,128,0.9848160147666931,0.976032018661499,1.0011520385742188,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:58,0.6.4
+sparse_multi_token_attention,torch,full,speed,ms,L,sequence length,256,0.9821439981460571,0.9684479832649231,1.0018240213394165,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:58,0.6.4
+sparse_multi_token_attention,torch,full,speed,ms,L,sequence length,512,1.0529279708862305,1.042956781387329,1.0710143804550172,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:58,0.6.4
+sparse_multi_token_attention,liger,backward,speed,ms,L,sequence length,32,0.2839040011167526,0.27542399168014525,0.2936640024185181,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:58,0.6.4
+sparse_multi_token_attention,liger,backward,speed,ms,L,sequence length,64,0.31360000371932983,0.30217599868774414,0.33344000577926636,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:58,0.6.4
+sparse_multi_token_attention,liger,backward,speed,ms,L,sequence length,128,0.3135039955377579,0.3035391986370087,0.3240256071090698,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:58,0.6.4
+sparse_multi_token_attention,liger,backward,speed,ms,L,sequence length,256,0.29872000217437744,0.2887359857559204,0.3113600015640259,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:58,0.6.4
+sparse_multi_token_attention,liger,backward,speed,ms,L,sequence length,512,0.7404800057411194,0.7379199862480164,0.7421375870704651,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:58,0.6.4
+sparse_multi_token_attention,torch,backward,speed,ms,L,sequence length,32,0.2640320062637329,0.2547904014587402,0.2865920066833496,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:59,0.6.4
+sparse_multi_token_attention,torch,backward,speed,ms,L,sequence length,64,0.28166401386260986,0.273062402009964,0.29402879476547245,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:59,0.6.4
+sparse_multi_token_attention,torch,backward,speed,ms,L,sequence length,128,0.2784639894962311,0.2717952013015747,0.28996479511260986,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:59,0.6.4
+sparse_multi_token_attention,torch,backward,speed,ms,L,sequence length,256,0.2667200118303299,0.2593280076980591,0.27438079118728637,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:59,0.6.4
+sparse_multi_token_attention,torch,backward,speed,ms,L,sequence length,512,0.42102399468421936,0.4184640049934387,0.42291200160980225,"{""B"": 2, ""C_in"": 4, ""C_out"": 4, ""K"": 3, ""groups"": 1, ""bias"": true, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:58:59,0.6.4
+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 HBM3,2025-12-19 00:58:59,0.6.4
+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 HBM3,2025-12-19 00:58:59,0.6.4
+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 HBM3,2025-12-19 00:58:59,0.6.4
+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 HBM3,2025-12-19 00:58:59,0.6.4
+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 HBM3,2025-12-19 00:58:59,0.6.4
+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 HBM3,2025-12-19 00:58:59,0.6.4
+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 HBM3,2025-12-19 00:58:59,0.6.4
+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 HBM3,2025-12-19 00:58:59,0.6.4
+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 HBM3,2025-12-19 00:58:59,0.6.4
+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 HBM3,2025-12-19 00:58:59,0.6.4
+sparsemax,liger,forward,speed,ms,V,feature size,1024,0.09596800059080124,0.09491200000047684,0.09681279957294464,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:08,0.6.4
+sparsemax,liger,forward,speed,ms,V,feature size,2048,0.18518400192260742,0.18409599363803864,0.18627199530601501,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:08,0.6.4
+sparsemax,liger,forward,speed,ms,V,feature size,4096,0.3548159897327423,0.35366401076316833,0.35596799850463867,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:08,0.6.4
+sparsemax,liger,forward,speed,ms,V,feature size,8192,1.55731201171875,1.5554239749908447,1.5587199926376343,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:08,0.6.4
+sparsemax,liger,forward,speed,ms,V,feature size,16384,3.0343360900878906,3.0321600437164307,3.03635835647583,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:08,0.6.4
+sparsemax,liger,forward,speed,ms,V,feature size,32768,6.036416053771973,6.034048080444336,6.0391998291015625,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:08,0.6.4
+sparsemax,torch,forward,speed,ms,V,feature size,1024,0.21491199731826782,0.21408000588417053,0.21587200462818146,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:11,0.6.4
+sparsemax,torch,forward,speed,ms,V,feature size,2048,0.39103999733924866,0.38995200395584106,0.3920319974422455,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:11,0.6.4
+sparsemax,torch,forward,speed,ms,V,feature size,4096,0.6908479928970337,0.6894720196723938,0.6921984195709228,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:11,0.6.4
+sparsemax,torch,forward,speed,ms,V,feature size,8192,2.1507198810577393,2.1486079692840576,2.152659273147583,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:11,0.6.4
+sparsemax,torch,forward,speed,ms,V,feature size,16384,4.0557920932769775,4.053343772888184,4.059040069580078,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:11,0.6.4
+sparsemax,torch,forward,speed,ms,V,feature size,32768,7.9557600021362305,7.953555297851563,7.958515262603759,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:11,0.6.4
+sparsemax,liger,full,speed,ms,V,feature size,1024,0.18411199748516083,0.1720000058412552,0.20998399853706362,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:16,0.6.4
+sparsemax,liger,full,speed,ms,V,feature size,2048,0.201664000749588,0.2009280025959015,0.20239999890327454,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:16,0.6.4
+sparsemax,liger,full,speed,ms,V,feature size,4096,0.39526399970054626,0.39420801401138306,0.39614080190658574,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:16,0.6.4
+sparsemax,liger,full,speed,ms,V,feature size,8192,1.6451200246810913,1.6406976222991942,1.6474303960800172,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:16,0.6.4
+sparsemax,liger,full,speed,ms,V,feature size,16384,3.197312116622925,3.195161628723145,3.1992640495300293,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:16,0.6.4
+sparsemax,liger,full,speed,ms,V,feature size,32768,6.438175916671753,6.4351487159729,6.4398334503173835,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:16,0.6.4
+sparsemax,torch,full,speed,ms,V,feature size,1024,0.3081440031528473,0.30170880556106566,0.3178432106971741,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:19,0.6.4
+sparsemax,torch,full,speed,ms,V,feature size,2048,0.5490239858627319,0.5466240048408508,0.5515199899673462,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:19,0.6.4
+sparsemax,torch,full,speed,ms,V,feature size,4096,1.0357120037078857,1.0330367803573608,1.039091205596924,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:19,0.6.4
+sparsemax,torch,full,speed,ms,V,feature size,8192,2.8234241008758545,2.8209919929504395,2.82586874961853,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:19,0.6.4
+sparsemax,torch,full,speed,ms,V,feature size,16384,5.3645758628845215,5.361625480651856,5.367948913574219,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:19,0.6.4
+sparsemax,torch,full,speed,ms,V,feature size,32768,10.52297592163086,10.519359588623047,10.5268798828125,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:19,0.6.4
+sparsemax,liger,backward,speed,ms,V,feature size,1024,0.05251200124621391,0.045343998819589615,0.06115199998021126,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:21,0.6.4
+sparsemax,liger,backward,speed,ms,V,feature size,2048,0.04944000020623207,0.043296001851558685,0.05881600081920624,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:21,0.6.4
+sparsemax,liger,backward,speed,ms,V,feature size,4096,0.04899200052022934,0.045471999794244766,0.05674879997968674,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:21,0.6.4
+sparsemax,liger,backward,speed,ms,V,feature size,8192,0.09657599776983261,0.09180799871683121,0.09772799909114838,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:21,0.6.4
+sparsemax,liger,backward,speed,ms,V,feature size,16384,0.16649599373340607,0.16540800034999847,0.16758400201797485,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:21,0.6.4
+sparsemax,liger,backward,speed,ms,V,feature size,32768,0.4017600119113922,0.4007999897003174,0.40278398990631104,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:21,0.6.4
+sparsemax,torch,backward,speed,ms,V,feature size,1024,0.10073599964380264,0.09510400146245956,0.10806400328874588,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:24,0.6.4
+sparsemax,torch,backward,speed,ms,V,feature size,2048,0.16595199704170227,0.16467200219631195,0.16709760427474976,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:24,0.6.4
+sparsemax,torch,backward,speed,ms,V,feature size,4096,0.3475840091705322,0.3463039994239807,0.3489919900894165,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:24,0.6.4
+sparsemax,torch,backward,speed,ms,V,feature size,8192,0.6755200028419495,0.6737599968910217,0.6771327733993531,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:24,0.6.4
+sparsemax,torch,backward,speed,ms,V,feature size,16384,1.3118400573730469,1.3104383707046507,1.313433575630188,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:24,0.6.4
+sparsemax,torch,backward,speed,ms,V,feature size,32768,2.565824031829834,2.563231945037842,2.5682880878448486,"{""B"": 4, ""T"": 512, ""dim"": -1, ""dtype"": ""torch.float32""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:24,0.6.4
+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 HBM3,2025-12-19 00:59:24,0.6.4
+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 HBM3,2025-12-19 00:59:24,0.6.4
+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 HBM3,2025-12-19 00:59:24,0.6.4
+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 HBM3,2025-12-19 00:59:24,0.6.4
+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 HBM3,2025-12-19 00:59:24,0.6.4
+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 HBM3,2025-12-19 00:59:24,0.6.4
+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 HBM3,2025-12-19 00:59:24,0.6.4
+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 HBM3,2025-12-19 00:59:24,0.6.4
+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 HBM3,2025-12-19 00:59:24,0.6.4
+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 HBM3,2025-12-19 00:59:24,0.6.4
+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 HBM3,2025-12-19 00:59:24,0.6.4
+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 HBM3,2025-12-19 00:59:24,0.6.4
+swiglu,liger,forward,speed,ms,T,sequence length,1024,1.5077760219573975,1.507468819618225,1.5093120336532593,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:34,0.6.4
+swiglu,liger,forward,speed,ms,T,sequence length,2048,2.9506560564041138,2.9493696689605713,2.9519424438476562,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:34,0.6.4
+swiglu,liger,forward,speed,ms,T,sequence length,4096,5.934688091278076,5.934688091278076,5.934688091278076,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:34,0.6.4
+swiglu,liger,forward,speed,ms,T,sequence length,8192,13.432831764221191,13.432831764221191,13.432831764221191,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:34,0.6.4
+swiglu,huggingface,forward,speed,ms,T,sequence length,1024,1.602832019329071,1.6017919778823853,1.6046080589294434,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:37,0.6.4
+swiglu,huggingface,forward,speed,ms,T,sequence length,2048,3.0784640312194824,3.0589568614959717,3.1841599464416506,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:37,0.6.4
+swiglu,huggingface,forward,speed,ms,T,sequence length,4096,6.187744140625,6.187744140625,6.187744140625,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:37,0.6.4
+swiglu,huggingface,forward,speed,ms,T,sequence length,8192,13.542943954467773,13.542943954467773,13.542943954467773,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:37,0.6.4
+swiglu,liger,full,speed,ms,T,sequence length,1024,5.04863977432251,5.04863977432251,5.04863977432251,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:40,0.6.4
+swiglu,liger,full,speed,ms,T,sequence length,2048,10.068191528320312,10.068191528320312,10.068191528320312,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:40,0.6.4
+swiglu,liger,full,speed,ms,T,sequence length,4096,20.635967254638672,20.635967254638672,20.635967254638672,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:40,0.6.4
+swiglu,liger,full,speed,ms,T,sequence length,8192,39.24220657348633,39.24220657348633,39.24220657348633,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:40,0.6.4
+swiglu,huggingface,full,speed,ms,T,sequence length,1024,5.058720111846924,5.058720111846924,5.058720111846924,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:43,0.6.4
+swiglu,huggingface,full,speed,ms,T,sequence length,2048,9.71513557434082,9.71513557434082,9.71513557434082,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:43,0.6.4
+swiglu,huggingface,full,speed,ms,T,sequence length,4096,21.262527465820312,21.262527465820312,21.262527465820312,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:43,0.6.4
+swiglu,huggingface,full,speed,ms,T,sequence length,8192,40.8928337097168,40.8928337097168,40.8928337097168,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:43,0.6.4
+swiglu,liger,backward,speed,ms,T,sequence length,1024,3.4040958881378174,3.4021566867828374,3.406035089492798,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:46,0.6.4
+swiglu,liger,backward,speed,ms,T,sequence length,2048,6.660128116607666,6.660128116607666,6.660128116607666,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:46,0.6.4
+swiglu,liger,backward,speed,ms,T,sequence length,4096,13.637696266174316,13.637696266174316,13.637696266174316,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:46,0.6.4
+swiglu,liger,backward,speed,ms,T,sequence length,8192,26.158815383911133,26.158815383911133,26.158815383911133,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:46,0.6.4
+swiglu,huggingface,backward,speed,ms,T,sequence length,1024,3.4464160203933716,3.4450815677642823,3.447750473022461,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:49,0.6.4
+swiglu,huggingface,backward,speed,ms,T,sequence length,2048,6.578688144683838,6.578688144683838,6.578688144683838,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:49,0.6.4
+swiglu,huggingface,backward,speed,ms,T,sequence length,4096,14.113951683044434,14.113951683044434,14.113951683044434,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:49,0.6.4
+swiglu,huggingface,backward,speed,ms,T,sequence length,8192,26.851648330688477,26.851648330688477,26.851648330688477,"{""B"": 4, ""hidden_size"": 4096, ""dtype"": ""torch.bfloat16"", ""intermediate_size"": 11008, ""hidden_act"": ""silu""}",NVIDIA H100 80GB HBM3,2025-12-19 00:59:49,0.6.4
+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 HBM3,2025-12-19 00:59:52,0.6.4
+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 HBM3,2025-12-19 00:59:52,0.6.4
+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 HBM3,2025-12-19 00:59:52,0.6.4
+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 HBM3,2025-12-19 00:59:52,0.6.4
+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 HBM3,2025-12-19 00:59:54,0.6.4
+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 HBM3,2025-12-19 00:59:54,0.6.4
+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 HBM3,2025-12-19 00:59:54,0.6.4
+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 HBM3,2025-12-19 00:59:54,0.6.4
+tiled_geglu,huggingface,full,speed,ms,T,sequence length,1024,0.5983040034770966,0.597599983215332,0.5993792176246643,"{""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 HBM3,2025-12-19 01:00:01,0.6.4
+tiled_geglu,huggingface,full,speed,ms,T,sequence length,2048,1.073039948940277,1.0722240209579468,1.0739519596099854,"{""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 HBM3,2025-12-19 01:00:01,0.6.4
+tiled_geglu,huggingface,full,speed,ms,T,sequence length,4096,2.025424003601074,2.0234111309051515,2.032902479171753,"{""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 HBM3,2025-12-19 01:00:01,0.6.4
+tiled_geglu,huggingface,full,speed,ms,T,sequence length,8192,3.958559989929199,3.957715129852295,3.9594048500061034,"{""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 HBM3,2025-12-19 01:00:01,0.6.4
+tiled_geglu,huggingface,full,speed,ms,T,sequence length,16384,7.896383762359619,7.896383762359619,7.896383762359619,"{""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 HBM3,2025-12-19 01:00:01,0.6.4
+tiled_geglu,liger,full,speed,ms,T,sequence length,1024,0.5567359924316406,0.5558592200279235,0.5581055998802185,"{""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 HBM3,2025-12-19 01:00:03,0.6.4
+tiled_geglu,liger,full,speed,ms,T,sequence length,2048,1.0085439682006836,1.0078208208084107,1.010150408744812,"{""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 HBM3,2025-12-19 01:00:03,0.6.4
+tiled_geglu,liger,full,speed,ms,T,sequence length,4096,1.880128026008606,1.878713607788086,1.8808704376220704,"{""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 HBM3,2025-12-19 01:00:03,0.6.4
+tiled_geglu,liger,full,speed,ms,T,sequence length,8192,3.7202080488204956,3.7192000389099125,3.7212160587310787,"{""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 HBM3,2025-12-19 01:00:03,0.6.4
+tiled_geglu,liger,full,speed,ms,T,sequence length,16384,7.57206392288208,7.57206392288208,7.57206392288208,"{""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 HBM3,2025-12-19 01:00:03,0.6.4
+tiled_geglu,liger_tiled,full,speed,ms,T,sequence length,1024,2.263535976409912,2.2415231227874757,2.2886464595794678,"{""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 HBM3,2025-12-19 01:00:04,0.6.4
+tiled_geglu,liger_tiled,full,speed,ms,T,sequence length,2048,2.290303945541382,2.278918409347534,2.350982332229614,"{""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 HBM3,2025-12-19 01:00:04,0.6.4
+tiled_geglu,liger_tiled,full,speed,ms,T,sequence length,4096,3.047136068344116,3.0452160358428957,3.0474048614501954,"{""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 HBM3,2025-12-19 01:00:04,0.6.4
+tiled_geglu,liger_tiled,full,speed,ms,T,sequence length,8192,5.832736015319824,5.832736015319824,5.832736015319824,"{""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 HBM3,2025-12-19 01:00:04,0.6.4
+tiled_geglu,liger_tiled,full,speed,ms,T,sequence length,16384,11.439264297485352,11.439264297485352,11.439264297485352,"{""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 HBM3,2025-12-19 01:00:04,0.6.4
+tiled_geglu,deepspeed_tiled,full,speed,ms,T,sequence length,1024,1.821951985359192,1.7989823818206787,1.850163221359253,"{""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 HBM3,2025-12-19 01:00:05,0.6.4
+tiled_geglu,deepspeed_tiled,full,speed,ms,T,sequence length,2048,1.9117760062217712,1.9090816020965575,1.9151359796524048,"{""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 HBM3,2025-12-19 01:00:05,0.6.4
+tiled_geglu,deepspeed_tiled,full,speed,ms,T,sequence length,4096,3.280832052230835,3.2799488544464115,3.281715250015259,"{""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 HBM3,2025-12-19 01:00:05,0.6.4
+tiled_geglu,deepspeed_tiled,full,speed,ms,T,sequence length,8192,6.020671844482422,6.020671844482422,6.020671844482422,"{""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 HBM3,2025-12-19 01:00:05,0.6.4
+tiled_geglu,deepspeed_tiled,full,speed,ms,T,sequence length,16384,11.948639869689941,11.948639869689941,11.948639869689941,"{""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 HBM3,2025-12-19 01:00:05,0.6.4
+tiled_geglu,huggingface,forward,speed,ms,T,sequence length,1024,0.17628799378871918,0.17578880190849305,0.1767359972000122,"{""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 HBM3,2025-12-19 01:00:06,0.6.4
+tiled_geglu,huggingface,forward,speed,ms,T,sequence length,2048,0.33446401357650757,0.33395839929580684,0.3354944050312042,"{""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 HBM3,2025-12-19 01:00:06,0.6.4
+tiled_geglu,huggingface,forward,speed,ms,T,sequence length,4096,0.6455039978027344,0.6451647758483887,0.6465728163719178,"{""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 HBM3,2025-12-19 01:00:06,0.6.4
+tiled_geglu,huggingface,forward,speed,ms,T,sequence length,8192,1.3082879781723022,1.306015968322754,1.308569598197937,"{""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 HBM3,2025-12-19 01:00:06,0.6.4
+tiled_geglu,huggingface,forward,speed,ms,T,sequence length,16384,2.5808000564575195,2.580070352554321,2.5817024230957033,"{""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 HBM3,2025-12-19 01:00:06,0.6.4
+tiled_geglu,liger,forward,speed,ms,T,sequence length,1024,0.16574399918317795,0.1644736021757126,0.16621440052986144,"{""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 HBM3,2025-12-19 01:00:06,0.6.4
+tiled_geglu,liger,forward,speed,ms,T,sequence length,2048,0.3103039860725403,0.30883840322494505,0.3117504060268402,"{""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 HBM3,2025-12-19 01:00:06,0.6.4
+tiled_geglu,liger,forward,speed,ms,T,sequence length,4096,0.6004959940910339,0.5986623883247375,0.601964795589447,"{""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 HBM3,2025-12-19 01:00:06,0.6.4
+tiled_geglu,liger,forward,speed,ms,T,sequence length,8192,1.2272000312805176,1.2256896018981933,1.2299200057983397,"{""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 HBM3,2025-12-19 01:00:06,0.6.4
+tiled_geglu,liger,forward,speed,ms,T,sequence length,16384,2.5393600463867188,2.5371327877044676,2.5403584003448487,"{""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 HBM3,2025-12-19 01:00:06,0.6.4
+tiled_geglu,liger_tiled,forward,speed,ms,T,sequence length,1024,0.4825280010700226,0.4776064038276673,0.48414719104766846,"{""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 HBM3,2025-12-19 01:00:07,0.6.4
+tiled_geglu,liger_tiled,forward,speed,ms,T,sequence length,2048,0.48287999629974365,0.4788671970367432,0.4870336055755615,"{""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 HBM3,2025-12-19 01:00:07,0.6.4
+tiled_geglu,liger_tiled,forward,speed,ms,T,sequence length,4096,0.7886719703674316,0.7879679799079895,0.7891200184822083,"{""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 HBM3,2025-12-19 01:00:07,0.6.4
+tiled_geglu,liger_tiled,forward,speed,ms,T,sequence length,8192,1.475663959980011,1.4738240242004395,1.4762879610061646,"{""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 HBM3,2025-12-19 01:00:07,0.6.4
+tiled_geglu,liger_tiled,forward,speed,ms,T,sequence length,16384,2.9471359252929688,2.946233558654785,2.9478271484375,"{""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 HBM3,2025-12-19 01:00:07,0.6.4
+tiled_geglu,deepspeed_tiled,forward,speed,ms,T,sequence length,1024,0.35233600437641144,0.34947839975357053,0.35971838831901554,"{""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 HBM3,2025-12-19 01:00:08,0.6.4
+tiled_geglu,deepspeed_tiled,forward,speed,ms,T,sequence length,2048,0.4721439927816391,0.4716480076313019,0.47303680181503294,"{""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 HBM3,2025-12-19 01:00:08,0.6.4
+tiled_geglu,deepspeed_tiled,forward,speed,ms,T,sequence length,4096,0.8339840173721313,0.8324288010597228,0.8357439994812011,"{""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 HBM3,2025-12-19 01:00:08,0.6.4
+tiled_geglu,deepspeed_tiled,forward,speed,ms,T,sequence length,8192,1.5999359488487244,1.5994240045547485,1.6010559797286987,"{""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 HBM3,2025-12-19 01:00:08,0.6.4
+tiled_geglu,deepspeed_tiled,forward,speed,ms,T,sequence length,16384,3.058720111846924,3.0578560829162598,3.059929609298706,"{""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 HBM3,2025-12-19 01:00:08,0.6.4
+tiled_geglu,huggingface,backward,speed,ms,T,sequence length,1024,0.4131839871406555,0.41240959167480473,0.41378558874130245,"{""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 HBM3,2025-12-19 01:00:09,0.6.4
+tiled_geglu,huggingface,backward,speed,ms,T,sequence length,2048,0.7178240120410919,0.7161471843719482,0.7197248220443726,"{""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 HBM3,2025-12-19 01:00:09,0.6.4
+tiled_geglu,huggingface,backward,speed,ms,T,sequence length,4096,1.3411200046539307,1.3405439853668213,1.3432960510253906,"{""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 HBM3,2025-12-19 01:00:09,0.6.4
+tiled_geglu,huggingface,backward,speed,ms,T,sequence length,8192,2.5897281169891357,2.5885568141937254,2.5912448883056642,"{""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 HBM3,2025-12-19 01:00:09,0.6.4
+tiled_geglu,huggingface,backward,speed,ms,T,sequence length,16384,5.416031837463379,5.416031837463379,5.416031837463379,"{""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 HBM3,2025-12-19 01:00:09,0.6.4
+tiled_geglu,liger,backward,speed,ms,T,sequence length,1024,0.3754560053348541,0.3746432065963745,0.3759423971176148,"{""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 HBM3,2025-12-19 01:00:10,0.6.4
+tiled_geglu,liger,backward,speed,ms,T,sequence length,2048,0.6598719954490662,0.6559743762016297,0.6608255863189697,"{""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 HBM3,2025-12-19 01:00:10,0.6.4
+tiled_geglu,liger,backward,speed,ms,T,sequence length,4096,1.2384320497512817,1.2361087560653687,1.2407488107681275,"{""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 HBM3,2025-12-19 01:00:10,0.6.4
+tiled_geglu,liger,backward,speed,ms,T,sequence length,8192,2.613759994506836,2.611859130859375,2.6139904499053954,"{""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 HBM3,2025-12-19 01:00:10,0.6.4
+tiled_geglu,liger,backward,speed,ms,T,sequence length,16384,4.965503931045532,4.965254306793213,4.965753555297852,"{""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 HBM3,2025-12-19 01:00:10,0.6.4
+tiled_geglu,liger_tiled,backward,speed,ms,T,sequence length,1024,1.5527520179748535,1.5311039686203003,1.5750720500946045,"{""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 HBM3,2025-12-19 01:00:11,0.6.4
+tiled_geglu,liger_tiled,backward,speed,ms,T,sequence length,2048,1.524176001548767,1.5012480020523071,1.5404800176620483,"{""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 HBM3,2025-12-19 01:00:11,0.6.4
+tiled_geglu,liger_tiled,backward,speed,ms,T,sequence length,4096,2.2332640886306763,2.23274884223938,2.233996868133545,"{""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 HBM3,2025-12-19 01:00:11,0.6.4
+tiled_geglu,liger_tiled,backward,speed,ms,T,sequence length,8192,3.9669920206069946,3.959552001953125,3.974432039260864,"{""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 HBM3,2025-12-19 01:00:11,0.6.4
+tiled_geglu,liger_tiled,backward,speed,ms,T,sequence length,16384,8.352895736694336,8.352895736694336,8.352895736694336,"{""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 HBM3,2025-12-19 01:00:11,0.6.4
+tiled_geglu,deepspeed_tiled,backward,speed,ms,T,sequence length,1024,1.192639946937561,1.1869184494018554,1.205836820602417,"{""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 HBM3,2025-12-19 01:00:11,0.6.4
+tiled_geglu,deepspeed_tiled,backward,speed,ms,T,sequence length,2048,1.419327974319458,1.4187840223312378,1.4208320379257202,"{""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 HBM3,2025-12-19 01:00:11,0.6.4
+tiled_geglu,deepspeed_tiled,backward,speed,ms,T,sequence length,4096,2.3672159910202026,2.3654719829559325,2.3674880027770997,"{""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 HBM3,2025-12-19 01:00:11,0.6.4
+tiled_geglu,deepspeed_tiled,backward,speed,ms,T,sequence length,8192,4.399376153945923,4.398118591308593,4.400633716583252,"{""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 HBM3,2025-12-19 01:00:11,0.6.4
+tiled_geglu,deepspeed_tiled,backward,speed,ms,T,sequence length,16384,8.470272064208984,8.470272064208984,8.470272064208984,"{""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 HBM3,2025-12-19 01:00:11,0.6.4
+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 HBM3,2025-12-19 01:00:12,0.6.4
+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 HBM3,2025-12-19 01:00:12,0.6.4
+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 HBM3,2025-12-19 01:00:12,0.6.4
+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 HBM3,2025-12-19 01:00:12,0.6.4
+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 HBM3,2025-12-19 01:00:12,0.6.4
+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 HBM3,2025-12-19 01:00:13,0.6.4
+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 HBM3,2025-12-19 01:00:13,0.6.4
+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 HBM3,2025-12-19 01:00:13,0.6.4
+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 HBM3,2025-12-19 01:00:13,0.6.4
+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 HBM3,2025-12-19 01:00:13,0.6.4
+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 HBM3,2025-12-19 01:00:13,0.6.4
+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 HBM3,2025-12-19 01:00:13,0.6.4
+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 HBM3,2025-12-19 01:00:13,0.6.4
+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 HBM3,2025-12-19 01:00:13,0.6.4
+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 HBM3,2025-12-19 01:00:13,0.6.4
+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 HBM3,2025-12-19 01:00:14,0.6.4
+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 HBM3,2025-12-19 01:00:14,0.6.4
+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 HBM3,2025-12-19 01:00:14,0.6.4
+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 HBM3,2025-12-19 01:00:14,0.6.4
+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 HBM3,2025-12-19 01:00:14,0.6.4
+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 HBM3,2025-12-19 01:00:15,0.6.4
+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 HBM3,2025-12-19 01:00:15,0.6.4
+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 HBM3,2025-12-19 01:00:15,0.6.4
+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 HBM3,2025-12-19 01:00:15,0.6.4
+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 HBM3,2025-12-19 01:00:15,0.6.4
+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 HBM3,2025-12-19 01:00:15,0.6.4
+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 HBM3,2025-12-19 01:00:15,0.6.4
+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 HBM3,2025-12-19 01:00:15,0.6.4
+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 HBM3,2025-12-19 01:00:15,0.6.4
+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 HBM3,2025-12-19 01:00:15,0.6.4
+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 HBM3,2025-12-19 01:00:16,0.6.4
+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 HBM3,2025-12-19 01:00:16,0.6.4
+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 HBM3,2025-12-19 01:00:16,0.6.4
+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 HBM3,2025-12-19 01:00:16,0.6.4
+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 HBM3,2025-12-19 01:00:16,0.6.4
+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 HBM3,2025-12-19 01:00:17,0.6.4
+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 HBM3,2025-12-19 01:00:17,0.6.4
+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 HBM3,2025-12-19 01:00:17,0.6.4
+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 HBM3,2025-12-19 01:00:17,0.6.4
+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 HBM3,2025-12-19 01:00:17,0.6.4
+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 HBM3,2025-12-19 01:00:17,0.6.4
+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 HBM3,2025-12-19 01:00:17,0.6.4
+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 HBM3,2025-12-19 01:00:17,0.6.4
+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 HBM3,2025-12-19 01:00:17,0.6.4
+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 HBM3,2025-12-19 01:00:17,0.6.4
+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 HBM3,2025-12-19 01:00:18,0.6.4
+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 HBM3,2025-12-19 01:00:18,0.6.4
+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 HBM3,2025-12-19 01:00:18,0.6.4
+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 HBM3,2025-12-19 01:00:18,0.6.4
+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 HBM3,2025-12-19 01:00:18,0.6.4
+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 HBM3,2025-12-19 01:00:18,0.6.4
+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 HBM3,2025-12-19 01:00:18,0.6.4
+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 HBM3,2025-12-19 01:00:18,0.6.4
+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 HBM3,2025-12-19 01:00:18,0.6.4
+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 HBM3,2025-12-19 01:00:18,0.6.4
+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 HBM3,2025-12-19 01:00:19,0.6.4
+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 HBM3,2025-12-19 01:00:19,0.6.4
+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 HBM3,2025-12-19 01:00:19,0.6.4
+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 HBM3,2025-12-19 01:00:19,0.6.4
+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 HBM3,2025-12-19 01:00:19,0.6.4
+tiled_swiglu,huggingface,full,speed,ms,T,sequence length,1024,0.5991359949111938,0.5988608121871948,0.5999231815338135,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:20,0.6.4
+tiled_swiglu,huggingface,full,speed,ms,T,sequence length,2048,1.0932000279426575,1.085651230812073,1.0941312074661256,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:20,0.6.4
+tiled_swiglu,huggingface,full,speed,ms,T,sequence length,4096,2.12992000579834,2.127891111373901,2.1308352947235107,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:20,0.6.4
+tiled_swiglu,huggingface,full,speed,ms,T,sequence length,8192,3.9650720357894897,3.9622208118438724,3.9679232597351075,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:20,0.6.4
+tiled_swiglu,huggingface,full,speed,ms,T,sequence length,16384,7.845632076263428,7.845632076263428,7.845632076263428,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:20,0.6.4
+tiled_swiglu,liger,full,speed,ms,T,sequence length,1024,0.5591999888420105,0.5577664136886596,0.5617920279502868,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:21,0.6.4
+tiled_swiglu,liger,full,speed,ms,T,sequence length,2048,1.0011520385742188,0.9985152125358582,1.045638370513916,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:21,0.6.4
+tiled_swiglu,liger,full,speed,ms,T,sequence length,4096,1.9300800561904907,1.8841471672058105,2.014636754989624,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:21,0.6.4
+tiled_swiglu,liger,full,speed,ms,T,sequence length,8192,3.9695839881896973,3.9694399356842043,3.9697280406951903,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:21,0.6.4
+tiled_swiglu,liger,full,speed,ms,T,sequence length,16384,7.7591681480407715,7.7591681480407715,7.7591681480407715,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:21,0.6.4
+tiled_swiglu,liger_tiled,full,speed,ms,T,sequence length,1024,2.3111519813537598,2.2866751194000243,2.3709824085235596,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:22,0.6.4
+tiled_swiglu,liger_tiled,full,speed,ms,T,sequence length,2048,2.315903902053833,2.3068031311035155,2.5785983085632322,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:22,0.6.4
+tiled_swiglu,liger_tiled,full,speed,ms,T,sequence length,4096,3.065824031829834,3.063808012008667,3.0662464618682863,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:22,0.6.4
+tiled_swiglu,liger_tiled,full,speed,ms,T,sequence length,8192,5.6030402183532715,5.6030402183532715,5.6030402183532715,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:22,0.6.4
+tiled_swiglu,liger_tiled,full,speed,ms,T,sequence length,16384,11.237183570861816,11.237183570861816,11.237183570861816,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:22,0.6.4
+tiled_swiglu,deepspeed_tiled,full,speed,ms,T,sequence length,1024,1.8142080307006836,1.7503104209899902,1.9010111808776855,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:23,0.6.4
+tiled_swiglu,deepspeed_tiled,full,speed,ms,T,sequence length,2048,1.917568027973175,1.9153664112091064,1.9204992055892944,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:23,0.6.4
+tiled_swiglu,deepspeed_tiled,full,speed,ms,T,sequence length,4096,3.265439987182617,3.2470848083496096,3.283795166015625,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:23,0.6.4
+tiled_swiglu,deepspeed_tiled,full,speed,ms,T,sequence length,8192,6.139071941375732,6.139071941375732,6.139071941375732,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:23,0.6.4
+tiled_swiglu,deepspeed_tiled,full,speed,ms,T,sequence length,16384,12.332863807678223,12.332863807678223,12.332863807678223,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:23,0.6.4
+tiled_swiglu,huggingface,forward,speed,ms,T,sequence length,1024,0.1767520010471344,0.17641599476337433,0.17729279398918152,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:24,0.6.4
+tiled_swiglu,huggingface,forward,speed,ms,T,sequence length,2048,0.3354559987783432,0.3341376066207886,0.33674879670143126,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:24,0.6.4
+tiled_swiglu,huggingface,forward,speed,ms,T,sequence length,4096,0.656063973903656,0.6486527919769287,0.6788351893424988,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:24,0.6.4
+tiled_swiglu,huggingface,forward,speed,ms,T,sequence length,8192,1.2930879592895508,1.292627215385437,1.2947520256042482,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:24,0.6.4
+tiled_swiglu,huggingface,forward,speed,ms,T,sequence length,16384,2.5672640800476074,2.5649216175079346,2.5675328731536866,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:24,0.6.4
+tiled_swiglu,liger,forward,speed,ms,T,sequence length,1024,0.16545599699020386,0.16407679915428164,0.16579840183258057,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:24,0.6.4
+tiled_swiglu,liger,forward,speed,ms,T,sequence length,2048,0.312063992023468,0.3104704022407532,0.3130112051963806,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:24,0.6.4
+tiled_swiglu,liger,forward,speed,ms,T,sequence length,4096,0.6011520028114319,0.5995840072631836,0.6029119968414307,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:24,0.6.4
+tiled_swiglu,liger,forward,speed,ms,T,sequence length,8192,1.2089600563049316,1.2075072288513184,1.229849624633789,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:24,0.6.4
+tiled_swiglu,liger,forward,speed,ms,T,sequence length,16384,2.389904022216797,2.3894144058227536,2.3912896156311034,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:24,0.6.4
+tiled_swiglu,liger_tiled,forward,speed,ms,T,sequence length,1024,0.4830079972743988,0.4814911961555481,0.49100799560546876,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:25,0.6.4
+tiled_swiglu,liger_tiled,forward,speed,ms,T,sequence length,2048,0.4887680113315582,0.4833791971206665,0.49119999408721926,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:25,0.6.4
+tiled_swiglu,liger_tiled,forward,speed,ms,T,sequence length,4096,0.7851200103759766,0.784608006477356,0.7862719893455505,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:25,0.6.4
+tiled_swiglu,liger_tiled,forward,speed,ms,T,sequence length,8192,1.4911519885063171,1.4900480508804321,1.4915519952774048,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:25,0.6.4
+tiled_swiglu,liger_tiled,forward,speed,ms,T,sequence length,16384,3.000704050064087,2.9462336063385006,3.0021248340606688,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:25,0.6.4
+tiled_swiglu,deepspeed_tiled,forward,speed,ms,T,sequence length,1024,0.3349759876728058,0.33157759308815005,0.33807359933853154,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:26,0.6.4
+tiled_swiglu,deepspeed_tiled,forward,speed,ms,T,sequence length,2048,0.4751040041446686,0.47371519207954405,0.47570560574531556,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:26,0.6.4
+tiled_swiglu,deepspeed_tiled,forward,speed,ms,T,sequence length,4096,0.831279993057251,0.8301760077476502,0.8319551825523377,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:26,0.6.4
+tiled_swiglu,deepspeed_tiled,forward,speed,ms,T,sequence length,8192,1.5821759700775146,1.582060766220093,1.5830080032348635,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:26,0.6.4
+tiled_swiglu,deepspeed_tiled,forward,speed,ms,T,sequence length,16384,3.0483200550079346,3.0461503982543947,3.0496256828308104,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:26,0.6.4
+tiled_swiglu,huggingface,backward,speed,ms,T,sequence length,1024,0.41488000750541687,0.4140863955020905,0.41512959599494936,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:27,0.6.4
+tiled_swiglu,huggingface,backward,speed,ms,T,sequence length,2048,0.7197760045528412,0.7183552145957947,0.7203840017318726,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:27,0.6.4
+tiled_swiglu,huggingface,backward,speed,ms,T,sequence length,4096,1.3356159925460815,1.332800030708313,1.3378560543060303,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:27,0.6.4
+tiled_swiglu,huggingface,backward,speed,ms,T,sequence length,8192,2.6679039001464844,2.6677887439727783,2.672416019439697,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:27,0.6.4
+tiled_swiglu,huggingface,backward,speed,ms,T,sequence length,16384,5.1178879737854,5.1178879737854,5.1178879737854,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:27,0.6.4
+tiled_swiglu,liger,backward,speed,ms,T,sequence length,1024,0.3742559999227524,0.3731648027896881,0.3752320110797882,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:28,0.6.4
+tiled_swiglu,liger,backward,speed,ms,T,sequence length,2048,0.6598879992961884,0.657369601726532,0.6623232007026673,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:28,0.6.4
+tiled_swiglu,liger,backward,speed,ms,T,sequence length,4096,1.2428480386734009,1.2407359838485719,1.2616064071655275,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:28,0.6.4
+tiled_swiglu,liger,backward,speed,ms,T,sequence length,8192,2.42409610748291,2.422425699234009,2.427609586715698,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:28,0.6.4
+tiled_swiglu,liger,backward,speed,ms,T,sequence length,16384,5.215199947357178,5.19292802810669,5.237471866607666,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:28,0.6.4
+tiled_swiglu,liger_tiled,backward,speed,ms,T,sequence length,1024,1.5487680435180664,1.5439039468765259,1.5830399990081787,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:29,0.6.4
+tiled_swiglu,liger_tiled,backward,speed,ms,T,sequence length,2048,1.527999997138977,1.5260800123214722,1.5574079751968384,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:29,0.6.4
+tiled_swiglu,liger_tiled,backward,speed,ms,T,sequence length,4096,2.2328479290008545,2.2316479206085207,2.2342271327972414,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:29,0.6.4
+tiled_swiglu,liger_tiled,backward,speed,ms,T,sequence length,8192,3.980080008506775,3.973177576065064,3.9869824409484864,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:29,0.6.4
+tiled_swiglu,liger_tiled,backward,speed,ms,T,sequence length,16384,7.56217622756958,7.56217622756958,7.56217622756958,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:29,0.6.4
+tiled_swiglu,deepspeed_tiled,backward,speed,ms,T,sequence length,1024,1.2047679424285889,1.1935487508773803,1.2256064414978027,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:29,0.6.4
+tiled_swiglu,deepspeed_tiled,backward,speed,ms,T,sequence length,2048,1.4278560280799866,1.4257919788360596,1.4298239946365356,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:29,0.6.4
+tiled_swiglu,deepspeed_tiled,backward,speed,ms,T,sequence length,4096,2.3844799995422363,2.382579135894775,2.3950783729553224,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:29,0.6.4
+tiled_swiglu,deepspeed_tiled,backward,speed,ms,T,sequence length,8192,4.270240068435669,4.270182418823243,4.270297718048096,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:29,0.6.4
+tiled_swiglu,deepspeed_tiled,backward,speed,ms,T,sequence length,16384,8.580863952636719,8.580863952636719,8.580863952636719,"{""bsz"": 2, ""hidden_size"": 2048, ""intermediate_size"": 4096, ""hidden_act"": ""silu"", ""activation_type"": ""swiglu"", ""num_shards"": 4, ""dtype"": ""torch.bfloat16""}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:29,0.6.4
+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 HBM3,2025-12-19 01:00:30,0.6.4
+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 HBM3,2025-12-19 01:00:30,0.6.4
+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 HBM3,2025-12-19 01:00:30,0.6.4
+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 HBM3,2025-12-19 01:00:30,0.6.4
+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 HBM3,2025-12-19 01:00:30,0.6.4
+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 HBM3,2025-12-19 01:00:31,0.6.4
+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 HBM3,2025-12-19 01:00:31,0.6.4
+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 HBM3,2025-12-19 01:00:31,0.6.4
+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 HBM3,2025-12-19 01:00:31,0.6.4
+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 HBM3,2025-12-19 01:00:31,0.6.4
+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 HBM3,2025-12-19 01:00:31,0.6.4
+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 HBM3,2025-12-19 01:00:31,0.6.4
+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 HBM3,2025-12-19 01:00:31,0.6.4
+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 HBM3,2025-12-19 01:00:31,0.6.4
+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 HBM3,2025-12-19 01:00:31,0.6.4
+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 HBM3,2025-12-19 01:00:32,0.6.4
+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 HBM3,2025-12-19 01:00:32,0.6.4
+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 HBM3,2025-12-19 01:00:32,0.6.4
+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 HBM3,2025-12-19 01:00:32,0.6.4
+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 HBM3,2025-12-19 01:00:32,0.6.4
+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 HBM3,2025-12-19 01:00:33,0.6.4
+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 HBM3,2025-12-19 01:00:33,0.6.4
+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 HBM3,2025-12-19 01:00:33,0.6.4
+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 HBM3,2025-12-19 01:00:33,0.6.4
+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 HBM3,2025-12-19 01:00:33,0.6.4
+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 HBM3,2025-12-19 01:00:33,0.6.4
+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 HBM3,2025-12-19 01:00:33,0.6.4
+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 HBM3,2025-12-19 01:00:33,0.6.4
+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 HBM3,2025-12-19 01:00:33,0.6.4
+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 HBM3,2025-12-19 01:00:33,0.6.4
+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 HBM3,2025-12-19 01:00:34,0.6.4
+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 HBM3,2025-12-19 01:00:34,0.6.4
+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 HBM3,2025-12-19 01:00:34,0.6.4
+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 HBM3,2025-12-19 01:00:34,0.6.4
+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 HBM3,2025-12-19 01:00:34,0.6.4
+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 HBM3,2025-12-19 01:00:35,0.6.4
+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 HBM3,2025-12-19 01:00:35,0.6.4
+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 HBM3,2025-12-19 01:00:35,0.6.4
+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 HBM3,2025-12-19 01:00:35,0.6.4
+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 HBM3,2025-12-19 01:00:35,0.6.4
+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 HBM3,2025-12-19 01:00:35,0.6.4
+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 HBM3,2025-12-19 01:00:35,0.6.4
+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 HBM3,2025-12-19 01:00:35,0.6.4
+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 HBM3,2025-12-19 01:00:35,0.6.4
+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 HBM3,2025-12-19 01:00:35,0.6.4
+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 HBM3,2025-12-19 01:00:36,0.6.4
+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 HBM3,2025-12-19 01:00:36,0.6.4
+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 HBM3,2025-12-19 01:00:36,0.6.4
+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 HBM3,2025-12-19 01:00:36,0.6.4
+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 HBM3,2025-12-19 01:00:36,0.6.4
+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 HBM3,2025-12-19 01:00:37,0.6.4
+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 HBM3,2025-12-19 01:00:37,0.6.4
+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 HBM3,2025-12-19 01:00:37,0.6.4
+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 HBM3,2025-12-19 01:00:37,0.6.4
+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 HBM3,2025-12-19 01:00:37,0.6.4
+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 HBM3,2025-12-19 01:00:37,0.6.4
+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 HBM3,2025-12-19 01:00:37,0.6.4
+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 HBM3,2025-12-19 01:00:37,0.6.4
+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 HBM3,2025-12-19 01:00:37,0.6.4
+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 HBM3,2025-12-19 01:00:37,0.6.4
+tvd,liger,full,memory,MB,V,vocab size,4096,1792.0009765625,1792.0009765625,1792.0009765625,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:45,0.6.4
+tvd,liger,full,memory,MB,V,vocab size,8192,3584.0009765625,3584.0009765625,3584.0009765625,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:45,0.6.4
+tvd,liger,full,memory,MB,V,vocab size,16384,7168.0009765625,7168.0009765625,7168.0009765625,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:45,0.6.4
+tvd,liger,full,memory,MB,V,vocab size,32768,14336.0009765625,14336.0009765625,14336.0009765625,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:45,0.6.4
+tvd,liger,full,memory,MB,V,vocab size,65536,28672.0,28672.0,28672.0,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:45,0.6.4
+tvd,liger,full,memory,MB,V,vocab size,131072,57344.0,57344.0,57344.0,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:45,0.6.4
+tvd,torch,full,memory,MB,V,vocab size,4096,2048.0009765625,2048.0009765625,2048.0009765625,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:45,0.6.4
+tvd,torch,full,memory,MB,V,vocab size,8192,4096.0009765625,4096.0009765625,4096.0009765625,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:45,0.6.4
+tvd,torch,full,memory,MB,V,vocab size,16384,8192.0009765625,8192.0009765625,8192.0009765625,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:45,0.6.4
+tvd,torch,full,memory,MB,V,vocab size,32768,16384.0,16384.0,16384.0,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:45,0.6.4
+tvd,torch,full,memory,MB,V,vocab size,65536,32768.0,32768.0,32768.0,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:45,0.6.4
+tvd,torch,full,memory,MB,V,vocab size,131072,65536.0,65536.0,65536.0,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:45,0.6.4
+tvd,liger,forward,speed,ms,V,vocab size,4096,0.46345600485801697,0.4622719883918762,0.46474880576133726,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:48,0.6.4
+tvd,liger,forward,speed,ms,V,vocab size,8192,0.903007984161377,0.901856005191803,0.9043520092964172,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:48,0.6.4
+tvd,liger,forward,speed,ms,V,vocab size,16384,1.7818880081176758,1.779871940612793,1.7829439640045166,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:48,0.6.4
+tvd,liger,forward,speed,ms,V,vocab size,32768,3.559040069580078,3.5574208736419677,3.5610175132751465,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:48,0.6.4
+tvd,liger,forward,speed,ms,V,vocab size,65536,7.157695770263672,7.155744171142578,7.158867263793945,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:48,0.6.4
+tvd,liger,forward,speed,ms,V,vocab size,131072,14.391472339630127,14.388640403747559,14.393600463867188,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:48,0.6.4
+tvd,torch,forward,speed,ms,V,vocab size,4096,0.7400959730148315,0.7394239902496338,0.7411199808120728,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:49,0.6.4
+tvd,torch,forward,speed,ms,V,vocab size,8192,1.4428800344467163,1.4415295600891114,1.443878412246704,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:49,0.6.4
+tvd,torch,forward,speed,ms,V,vocab size,16384,2.852239966392517,2.851225566864014,2.8547200679779055,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:49,0.6.4
+tvd,torch,forward,speed,ms,V,vocab size,32768,5.659135818481445,5.657132720947265,5.66033296585083,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:49,0.6.4
+tvd,torch,forward,speed,ms,V,vocab size,65536,11.307712078094482,11.30648307800293,11.310623931884766,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:49,0.6.4
+tvd,torch,forward,speed,ms,V,vocab size,131072,22.611424446105957,22.605338287353515,22.616588973999022,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:49,0.6.4
+tvd,liger,full,speed,ms,V,vocab size,4096,1.333888053894043,1.3266879796981812,1.353663969039917,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:50,0.6.4
+tvd,liger,full,speed,ms,V,vocab size,8192,2.5678720474243164,2.5607295989990235,2.5909247398376465,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:50,0.6.4
+tvd,liger,full,speed,ms,V,vocab size,16384,5.337424039840698,5.3342592239379885,5.350246334075928,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:50,0.6.4
+tvd,liger,full,speed,ms,V,vocab size,32768,10.635871887207031,10.632838439941406,10.640755462646485,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:50,0.6.4
+tvd,liger,full,speed,ms,V,vocab size,65536,21.36415958404541,21.341382217407226,21.430022430419925,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:50,0.6.4
+tvd,liger,full,speed,ms,V,vocab size,131072,42.76108932495117,42.759706878662115,42.76247177124023,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:50,0.6.4
+tvd,torch,full,speed,ms,V,vocab size,4096,2.153424024581909,2.151955223083496,2.155161666870117,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:52,0.6.4
+tvd,torch,full,speed,ms,V,vocab size,8192,4.246175765991211,4.245228672027588,4.247648048400879,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:52,0.6.4
+tvd,torch,full,speed,ms,V,vocab size,16384,8.753567695617676,8.751680374145508,8.755200386047363,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:52,0.6.4
+tvd,torch,full,speed,ms,V,vocab size,32768,17.472959518432617,17.472742080688477,17.47750358581543,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:52,0.6.4
+tvd,torch,full,speed,ms,V,vocab size,65536,35.00495910644531,35.004709625244146,35.005208587646486,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:52,0.6.4
+tvd,torch,full,speed,ms,V,vocab size,131072,70.09337615966797,70.09337615966797,70.09337615966797,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:52,0.6.4
+tvd,liger,backward,speed,ms,V,vocab size,4096,0.8568000197410583,0.8528000116348267,0.8655359864234924,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:53,0.6.4
+tvd,liger,backward,speed,ms,V,vocab size,8192,1.6576639413833618,1.6485504388809205,1.6695488214492797,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:53,0.6.4
+tvd,liger,backward,speed,ms,V,vocab size,16384,3.5667519569396973,3.5488255500793455,3.6032384395599366,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:53,0.6.4
+tvd,liger,backward,speed,ms,V,vocab size,32768,7.094304084777832,7.07501449584961,7.147903919219971,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:53,0.6.4
+tvd,liger,backward,speed,ms,V,vocab size,65536,14.164671897888184,14.152447700500488,14.187423706054688,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:53,0.6.4
+tvd,liger,backward,speed,ms,V,vocab size,131072,28.34841537475586,28.345304870605467,28.349587249755857,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:53,0.6.4
+tvd,torch,backward,speed,ms,V,vocab size,4096,1.4163199663162231,1.4150591611862182,1.4179199934005737,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:54,0.6.4
+tvd,torch,backward,speed,ms,V,vocab size,8192,2.8075520992279053,2.806476879119873,2.8092545032501217,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:54,0.6.4
+tvd,torch,backward,speed,ms,V,vocab size,16384,5.908351898193359,5.906047821044922,5.9106879234313965,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:54,0.6.4
+tvd,torch,backward,speed,ms,V,vocab size,32768,11.818960189819336,11.815244483947755,11.821363067626955,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:54,0.6.4
+tvd,torch,backward,speed,ms,V,vocab size,65536,23.70097541809082,23.70032653808594,23.701727294921874,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:54,0.6.4
+tvd,torch,backward,speed,ms,V,vocab size,131072,47.523502349853516,47.52285919189453,47.5241455078125,"{""B"": 8, ""T"": 2048}",NVIDIA H100 80GB HBM3,2025-12-19 01:00:54,0.6.4
diff --git a/benchmarks/commits.txt b/benchmarks/commits.txt
index 38e65414..4bb82072 100644
--- a/benchmarks/commits.txt
+++ b/benchmarks/commits.txt
@@ -19,3 +19,4 @@ a80f30b
 0a62700
 3d075fd
 49e6353
+6059dfb