==1223237== Profiling result:
            Type  Time(%)      Time     Calls       Avg       Min       Max  Name
 GPU activities:   16.08%  15.1883s     83954  180.91us     256ns  92.642ms  [CUDA memcpy HtoD]
                   11.51%  10.8694s      1152  9.4352ms  200.55us  103.18ms  maxwell_sgemm_fp16_64x64_tn
                   10.96%  10.3471s       648  15.968ms  300.33us  65.592ms  maxwell_sgemm_fp16_32x128_tn
                    5.31%  5.01486s     14965  335.11us  25.217us  768.42us  void attn_chunk_kernel_pre75<int=1, int=2, bool=0>(float const *, strata::kernels::QsaAttnPools, int const *, int const , int, int, float, float*, int const *, int const *, int, int, __int64)
                    3.46%  3.26976s      3266  1.0012ms  43.841us  5.4238ms  void mul_mat_q<ggml_type=20, int=64, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    2.88%  2.71582s       768  3.5362ms  93.220us  9.2698ms  void dequant_kernel<int=14, H16>(unsigned char const *, long, long, long, long, long, H16*)
                    2.86%  2.70045s      1655  1.6317ms  178.76us  7.7256ms  void mul_mat_q<ggml_type=22, int=64, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    2.50%  2.36509s     23827  99.261us  8.0960us  2.1880ms  void native_q6_k_mmvq_kernel<bool=0>(Q6KBlock const *, Q81Block const *, float*, int, int)
                    2.40%  2.26797s      1443  1.5717ms  171.05us  8.1444ms  void mul_mat_q<ggml_type=18, int=64, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    2.32%  2.19528s       783  2.8037ms  464.18us  41.079ms  sgemm_128x128x8_NT_vec
                    2.08%  1.96923s      1527  1.2896ms  194.35us  1.8325ms  maxwell_sgemm_128x64_tn
                    1.84%  1.74172s     35880  48.542us  2.3680us  246.70us  void native_down_multi_kernel<int=20, bool=1>(__int64 const *, int const *, int const , int const , block_q8_1 const *, strata::kernels::NativeExpertLayout, float*)
                    1.51%  1.42466s     17415  81.806us  8.3520us  2.6303ms  void native_mmvq_multi_kernel<Q6KTraits, int=2, int=4, int=2, bool=0, bool=0>(Q6KTraitsBlock const *, Q81Block const *, float*, int, int, Q6KTraitsBlock const , Q81Block const *)
                    1.47%  1.38727s     15640  88.699us  2.4960us  421.49us  void native_gu_multi_kernel<int=18, bool=0, bool=1>(__int64 const *, int const *, int const , int const , block_q8_1 const *, strata::kernels::NativeExpertLayout, float*, strata::kernels::NativeExpertLayout)
                    1.46%  1.37823s       847  1.6272ms  85.635us  6.2062ms  void mul_mat_q<ggml_type=21, int=64, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    1.38%  1.30607s      8715  149.86us  2.0160us  553.62us  bf16_to_f32_kernel(unsigned short const *, float*, long)
                    1.34%  1.26231s       216  5.8440ms  806.40us  14.335ms  gdn_rec_cols_pipe_kernel(float*, float const *, float const *, float const *, float*, long)
                    1.31%  1.23884s     18400  67.328us  2.4960us  318.57us  void native_gu_multi_kernel<int=22, bool=1, bool=1>(__int64 const *, int const *, int const , int const , block_q8_1 const *, strata::kernels::NativeExpertLayout, float*, strata::kernels::NativeExpertLayout)
                    1.22%  1.15237s     66240  17.396us  3.0400us  13.483ms  wait_flag_ge_kernel(unsigned int volatile *, unsigned int)
                    1.14%  1.07248s     10965  97.809us  11.521us  3.1190ms  void native_mmvq_multi_kernel<Q6KTraits, int=3, int=4, int=2, bool=0, bool=0>(Q6KTraitsBlock const *, Q81Block const *, float*, int, int, Q6KTraitsBlock const , Q81Block const *)
                    1.07%  1.00929s     14965  67.443us  8.8000us  288.49us  attn_merge_kernel(float const *, float const *, float const *, int, float*, __int64)
                    0.95%  901.77ms      9200  98.018us  2.4960us  449.91us  void native_gu_multi_kernel<int=21, bool=0, bool=1>(__int64 const *, int const *, int const , int const , block_q8_1 const *, strata::kernels::NativeExpertLayout, float*, strata::kernels::NativeExpertLayout)
                    0.95%  893.97ms      3063  291.86us  58.691us  826.76us  sgemm_32x32x32_NT_vec
                    0.87%  818.43ms      7224  113.29us  12.929us  3.7616ms  void native_mmvq_multi_kernel<Q6KTraits, int=4, int=4, int=2, bool=0, bool=0>(Q6KTraitsBlock const *, Q81Block const *, float*, int, int, Q6KTraitsBlock const , Q81Block const *)
                    0.86%  813.56ms       318  2.5584ms  375.60us  5.0059ms  gr_write_norm_rs_kernel(float*, float const *, float const *, long, float const *, float, float*, unsigned short*, unsigned short*, long)
                    0.81%  765.62ms       585  1.3087ms  190.44us  2.5104ms  gr_mix_r_kernel(float const *, float const *, float const *, float const *, float*, unsigned short*, long, unsigned short*, unsigned short*)
                    0.80%  753.09ms       736  1.0232ms  101.83us  4.8352ms  void mul_mat_q<ggml_type=42, int=64, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.73%  688.81ms     14891  46.256us  7.2650us  267.60us  copy16_group_kernel(GroupArgs, int, long, long, uint4*, uint4*)
                    0.68%  643.54ms     19644  32.760us  28.769us  62.243us  void gr_down_staged_kernel<int=1, bool=1>(GrMulti)
                    0.67%  632.89ms      4743  133.44us  2.0170us  3.1120ms  void quantize_mmq_q8_1<mmq_q8_1_ds_layout=0, bool=0>(float const *, int const *, void*, long, long, long, long, long, int, int, int)
                    0.61%  574.43ms     19646  29.239us  26.913us  55.234us  void gr_up_multi_kernel<int=1, bool=1>(GrMulti)
                    0.56%  526.06ms     13832  38.031us  33.602us  67.779us  void gr_down_staged_kernel<int=2, bool=1>(GrMulti)
                    0.49%  463.64ms     19106  24.266us  8.8000us  2.2939ms  cvec_kernel(float*, float const *, float const *, int const *, int, long, int, int, long, float const *, long, float const *, long, int)
                    0.48%  453.68ms       288  1.5753ms  232.91us  3.0165ms  moe_combine4_kernel(float const *, int const *, float const *, float4 const *, float const *, float4*, long)
                    0.48%  451.28ms     13246  34.069us  28.833us  49.378us  void gr_up_multi_kernel<int=2, bool=1>(GrMulti)
                    0.39%  370.36ms     47253  7.8370us  6.7200us  25.633us  gr_norm_split_kernel(GrMulti)
                    0.37%  350.74ms       258  1.3595ms  198.12us  2.7444ms  gr_write_kernel(float*, float const *, float const *, long, long)
                    0.36%  341.39ms      8280  41.230us  3.4560us  178.31us  void native_down_multi_kernel<int=42, bool=1>(__int64 const *, int const *, int const , int const , block_q8_1 const *, strata::kernels::NativeExpertLayout, float*)
                    0.36%  337.55ms      8303  40.653us  35.874us  66.883us  void gr_down_staged_kernel<int=3, bool=1>(GrMulti)
                    0.35%  333.23ms      8305  40.124us  36.386us  49.538us  void gr_up_multi_kernel<int=3, bool=1>(GrMulti)
                    0.35%  326.02ms     22038  14.793us  5.4410us  20.833us  copy16_kernel(uint4 const *, long, uint4 const *, long, uint4*, uint4 const *, long, uint4*)
                    0.32%  303.13ms      1116  271.62us  10.592us  710.37us  block_scores_kernel(float const *, float const *, float const *, int const *, long, float*)
                    0.32%  297.68ms       267  1.1149ms  161.80us  2.4074ms  gr_norm_rs_kernel(float const *, float const *, float, float*, unsigned short*, unsigned short*, long)
                    0.31%  290.21ms      4419  65.672us  2.0480us  511.51us  swiglu_kernel(float const *, float*, long, long, bool)
                    0.29%  275.61ms      6054  45.524us  38.946us  72.355us  void gr_down_staged_kernel<int=4, bool=1>(GrMulti)
                    0.29%  274.67ms      6230  44.089us  7.6160us  90.628us  void native_q5_k_mmvq_kernel<bool=0>(Q5KBlock const *, Q81Block const *, float*, int, int)
                    0.29%  271.89ms     10152  26.781us  22.593us  40.513us  void gdn_step_norm_multi_kernel<bool=1, bool=0>(float*, float const *, int, float const , float const , float const , float const , float, gdn_step_norm_multi_kernel<bool=1, bool=0>, int, int, int, int const *, int, GdnQ81*)
                    0.28%  269.04ms      4725  56.940us  9.5360us  117.00us  void native_mmvq_multi_kernel<Q5KTraits, int=2, int=4, int=2, bool=0, bool=0>(Q5KTraitsBlock const *, Q81Block const *, float*, int, int, Q5KTraitsBlock const , Q81Block const *)
                    0.28%  263.98ms     10152  26.002us  17.537us  53.667us  void gdn_step_commit_kernel<bool=0>(float*, float const *, int, float const , float const , int, int, int const *, int)
                    0.28%  262.43ms      7476  35.103us  9.1520us  127.97us  void native_iq4_xs_mmvq_kernel<bool=1>(IQ4XSBlock const *, Q81Block const *, float*, int, int)
                    0.27%  257.71ms     13536  19.038us  18.464us  19.745us  route_multi(float const *, int*, float*, int)
                    0.27%  257.10ms      5460  47.087us  42.978us  57.282us  void gr_up_multi_kernel<int=4, bool=1>(GrMulti)
                    0.27%  250.36ms     23446  10.678us  2.9120us  173.38us  void bf16_f32_mmvf_kernel<int=256>(float const *, unsigned short const *, float*, int)
                    0.26%  242.97ms       216  1.1249ms  162.66us  2.0993ms  gdn_conv_tiled_kernel(float const *, float const *, float const *, float*, long)
                    0.23%  213.21ms      5392  39.542us  8.0000us  141.61us  void native_small_mmvq_kernel<Q80Block, int=8, bool=0>(Q80Block const *, Q81Block const *, float*, int, int)
                    0.22%  211.22ms      2975  70.999us  11.232us  147.27us  void native_mmvq_multi_kernel<Q5KTraits, int=3, int=4, int=2, bool=0, bool=0>(Q5KTraitsBlock const *, Q81Block const *, float*, int, int, Q5KTraitsBlock const , Q81Block const *)
                    0.22%  204.15ms      5670  36.005us  10.721us  128.42us  void native_mmvq_multi_kernel<IQ4XSTraits, int=2, int=4, int=2, bool=0, bool=0>(IQ4XSTraitsBlock const *, Q81Block const *, float*, int, int, IQ4XSTraitsBlock const , Q81Block const *)
                    0.21%  194.83ms     16470  11.829us  3.3280us  207.08us  void bf16_f32_mmvf_multi_kernel<int=256, int=2, bool=1>(float const *, long, unsigned short const *, float*, long, int, int)
                    0.20%  192.60ms      9409  20.469us  18.689us  64.099us  route(float const *, int*, float*)
                    0.20%  190.54ms     72217  2.6380us  1.8240us  13.569us  native_quantize_q8_1_kernel(float const *, Q81Block*, int)
                    0.20%  189.65ms      6345  29.889us  8.7690us  102.50us  void native_mmvq_multi_kernel<Q4KTraits, int=2, int=4, int=2, bool=0, bool=0>(Q4KTraitsBlock const *, Q81Block const *, float*, int, int, Q4KTraitsBlock const , Q81Block const *)
                    0.19%  181.85ms      8366  21.736us  6.8480us  73.475us  void native_q4_k_mmvq_kernel<bool=0>(Q4KBlock const *, Q81Block const *, float*, int, int)
                    0.19%  181.38ms      1960  92.538us  14.945us  199.72us  void native_mmvq_multi_kernel<Q5KTraits, int=4, int=4, int=2, bool=0, bool=0>(Q5KTraitsBlock const *, Q81Block const *, float*, int, int, Q5KTraitsBlock const , Q81Block const *)
                    0.19%  177.26ms      1027  172.60us     768ns  11.505ms  [CUDA memcpy DtoH]
                    0.19%  177.22ms      8366  21.183us  19.009us  26.785us  void native_small_mmvq_kernel<IQ4NLBlock, int=4, bool=1>(IQ4NLBlock const *, Q81Block const *, float*, int, int)
                    0.18%  174.48ms       216  807.79us  127.59us  1.5118ms  gdn_out_norm_kernel(float const *, float const *, float, float*, unsigned short*, long)
                    0.17%  156.96ms      3995  39.288us  10.497us  139.46us  void native_mmvq_multi_kernel<Q4KTraits, int=3, int=4, int=2, bool=0, bool=0>(Q4KTraitsBlock const *, Q81Block const *, float*, int, int, Q4KTraitsBlock const , Q81Block const *)
                    0.16%  150.29ms      3570  42.097us  11.105us  147.91us  void native_mmvq_multi_kernel<IQ4XSTraits, int=3, int=4, int=2, bool=0, bool=0>(IQ4XSTraitsBlock const *, Q81Block const *, float*, int, int, IQ4XSTraitsBlock const , Q81Block const *)
                    0.16%  149.33ms       210  711.08us  73.347us  1.6177ms  void dequant_kernel<int=13, H16>(unsigned char const *, long, long, long, long, long, H16*)
                    0.16%  147.13ms     10370  14.188us  3.8080us  236.62us  void bf16_f32_mmvf_multi_kernel<int=256, int=3, bool=1>(float const *, long, unsigned short const *, float*, long, int, int)
                    0.15%  145.45ms       282  515.77us  106.41us  1.9985ms  void dequant_kernel<int=12, H16>(unsigned char const *, long, long, long, long, long, H16*)
                    0.15%  142.85ms     44160  3.2340us  1.6640us  8.2240us  swiglu_q8_1_entries_kernel(float const *, float const *, int const *, int const *, int, block_q8_1*)
                    0.15%  136.99ms     11050  12.397us     288ns  53.049ms  [CUDA memset]
                    0.14%  132.99ms     22080  6.0220us  3.0720us  37.121us  strata::kernels::copy_rows_from_mapped_kernel(float4*, float4 volatile *, long, int const *, int const *, unsigned int volatile *, unsigned int)
                    0.14%  132.01ms     22080  5.9780us  4.3200us  21.313us  strata::kernels::doorbell_publish_res_kernel(float const *, int const *, int const *, int, int, int, float*, int*, unsigned int*)
                    0.14%  130.80ms     23742  5.5090us  1.8560us  21.153us  strata::kernels::copy_i32_from_mapped_kernel(int*, int volatile *, int)
                    0.14%  128.30ms      2632  48.746us  12.353us  172.74us  void native_mmvq_multi_kernel<Q4KTraits, int=4, int=4, int=2, bool=0, bool=0>(Q4KTraitsBlock const *, Q81Block const *, float*, int, int, Q4KTraitsBlock const , Q81Block const *)
                    0.13%  126.34ms      6408  19.715us  18.593us  21.313us  void gdn_step_norm_multi_kernel<bool=0, bool=0>(float*, float const *, int, float const , float const , float const , float const , float, gdn_step_norm_multi_kernel<bool=0, bool=0>, int, int, int, int const *, int, GdnQ81*)
                    0.13%  124.09ms      6345  19.556us  18.528us  24.353us  void native_mmvq_multi_kernel<SmallTraits<IQ4NLBlock, int=4>, int=2, int=4, int=2, bool=0, bool=0>(IQ4NLBlockBlock const *, Q81Block const *, float*, int, int, IQ4NLBlockBlock const , Q81Block const *)
                    0.13%  123.47ms       774  159.52us  15.584us  5.2336ms  void native_mmvq_multi_kernel<Q6KTraits, int=6, int=4, int=2, bool=0, bool=0>(Q6KTraitsBlock const *, Q81Block const *, float*, int, int, Q6KTraitsBlock const , Q81Block const *)
                    0.13%  119.44ms        57  2.0955ms  232.87us  8.2103ms  void mul_mat_q<ggml_type=23, int=64, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.12%  115.53ms      2352  49.120us  13.440us  171.11us  void native_mmvq_multi_kernel<IQ4XSTraits, int=4, int=4, int=2, bool=0, bool=0>(IQ4XSTraitsBlock const *, Q81Block const *, float*, int, int, IQ4XSTraitsBlock const , Q81Block const *)
                    0.11%  105.97ms      6832  15.510us  4.9600us  302.93us  void bf16_f32_mmvf_multi_kernel<int=256, int=4, bool=1>(float const *, long, unsigned short const *, float*, long, int, int)
                    0.11%  104.25ms        36  2.8959ms  137.86us  21.939ms  void mul_mat_q<ggml_type=8, int=64, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.10%  98.225ms       252  389.78us  81.444us  1.5700ms  void dequant_kernel<int=23, H16>(unsigned char const *, long, long, long, long, long, H16*)
                    0.10%  96.556ms       243  397.35us  13.920us  2.5645ms  rms_rows_kernel(float*, float const *, long, long, float)
                    0.10%  93.514ms       216  432.94us  65.731us  834.95us  gdn_l2_kernel(float*, float)
                    0.09%  86.435ms       920  93.951us  2.7520us  374.06us  void native_gu_multi_kernel<int=23, bool=0, bool=1>(__int64 const *, int const *, int const , int const , block_q8_1 const *, strata::kernels::NativeExpertLayout, float*, strata::kernels::NativeExpertLayout)
                    0.09%  86.198ms      3995  21.576us  20.320us  24.961us  void native_mmvq_multi_kernel<SmallTraits<IQ4NLBlock, int=4>, int=3, int=4, int=2, bool=0, bool=0>(IQ4NLBlockBlock const *, Q81Block const *, float*, int, int, IQ4NLBlockBlock const , Q81Block const *)
                    0.08%  80.147ms      1116  71.816us  1.4720us  157.74us  block_topk_kernel(float const *, int const *, long, long, int*)
                    0.08%  76.701ms     16560  4.6310us  4.0320us  6.1130us  gdn_conv_l2_multi_kernel(float*, float const *, float const *, float*, int, int, float, int, bool)
                    0.08%  75.984ms     22080  3.4410us  1.9520us  6.4330us  add_hits_kernel(float*, float const *, int const *, int const *, int)
                    0.08%  73.100ms      1325  55.169us  38.082us  91.940us  sampler_greedy_kernel(float const *, int, int const *, int, strata::kernels::SamplerParams, int, int, int*)
                    0.08%  72.744ms     22945  3.1700us  2.8480us  10.048us  combine_k10_vec4(float4 const *, float const *, float4 const *, float4*, long)
                    0.07%  70.310ms        72  976.53us  144.13us  1.9930ms  gate_attn_kernel(float const *, float const *, unsigned short*, long, long)
                    0.07%  69.594ms     22945  3.0330us  2.2720us  9.9210us  native_swiglu_quantize_q8_1_kernel(float const *, float const *, Q81Block*, int)
                    0.07%  69.070ms        81  852.72us  124.39us  1.9369ms  kv_append_kernel(float const *, float const *, long, int const *, long, unsigned short*, unsigned short*, char*, char*, unsigned short*, unsigned short*, strata::kernels::KvHostPools, strata::kernels::KvHostPools)
                    0.07%  65.188ms     18069  3.6070us  2.4640us  12.256us  void norm_rope_kernel<bool=0>(float const *, int, float const , float*, int, int, float, float, float, float, float, float, float, int const *, int const , strata::kernels::RopeTab)
                    0.07%  64.725ms     22080  2.9310us  2.6240us  4.0320us  quantize_q8_1_kernel(float const *, block_q8_1*, __int64)
                    0.07%  62.977ms      5520  11.408us  2.0160us  50.530us  block_topk_wide_kernel(float const *, int const *, long, long, int*)
                    0.07%  62.592ms      2632  23.781us  22.977us  26.145us  void native_mmvq_multi_kernel<SmallTraits<IQ4NLBlock, int=4>, int=4, int=4, int=2, bool=0, bool=0>(IQ4NLBlockBlock const *, Q81Block const *, float*, int, int, IQ4NLBlockBlock const , Q81Block const *)
                    0.06%  60.525ms     22080  2.7410us  2.4960us  3.6160us  fetch_blobs_kernel(__int64 const *, int const *, uint4*, __int64)
                    0.06%  56.291ms     22945  2.4530us  1.6960us  8.5440us  sigmoid_scale_rows_vec4_kernel(float4*, float const *, int)
                    0.06%  56.151ms        72  779.87us  108.13us  1.8545ms  split_q_kernel(float const *, float*, long)
                    0.06%  52.257ms      5520  9.4660us  2.5600us  416.85us  copy_kernel(strata::kernels::KvStreamMap, Runs)
                    0.05%  48.225ms      5520  8.7360us  4.6400us  26.625us  resolve_kernel(strata::kernels::KvStreamMap, int const *, int const *, int, int, int)
                    0.05%  45.454ms       594  76.522us  68.995us  96.804us  void gr_up_multi_kernel<int=6, bool=1>(GrMulti)
                    0.05%  45.005ms      8904  5.0540us  2.8480us  29.122us  void append<bool=0>(float const *, int const *, int, int, int, float const , float, float*, int const *, int const *, int*, int, float, float, float, float, float, float, int const , strata::kernels::RopeTab)
                    0.05%  44.243ms      6391  6.9220us  5.7600us  23.137us  void kv_append_q8_kernel<bool=1>(char*, kv_append_q8_kernel<bool=1>, unsigned short*, char*, int const *, int const , float const *, float const , int, int, int, strata::kernels::KvHostPools, int, int)
                    0.05%  42.842ms        93  460.66us  85.027us  643.87us  void mul_mat_q<ggml_type=22, int=48, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.05%  42.821ms       170  251.89us  80.676us  348.46us  void mul_mat_q<ggml_type=20, int=48, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.04%  42.429ms       432  98.215us  25.249us  241.35us  void gemv2T_kernel_val<int, int, float, float, float, float, int=128, int=16, int=2, int=2, bool=0, bool=0, cublasGemvParamsEx<int, cublasGemvTensorStridedBatched<float const >, cublasGemvTensorStridedBatched<float const >, cublasGemvTensorStridedBatched<float>, float>>(int, float, float)
                    0.04%  42.074ms     10152  4.1440us  2.6240us  16.673us  gdn_conv_commit_kernel(float*, float const *, int, int const *)
                    0.04%  41.542ms      4860  8.5470us  8.1600us  9.0570us  void gdn_ab_multi_kernel<int=2, bool=1>(float const *, unsigned short const *, unsigned short const , float const , float const , float*, unsigned short const *, int, int, int)
                    0.04%  41.389ms       260  159.19us  4.1040us  921.39us  [CUDA memcpy HtoH]
                    0.04%  41.151ms      6408  6.4210us  6.0800us  6.9760us  void gdn_ab_multi_kernel<int=1, bool=1>(float const *, unsigned short const *, unsigned short const , float const , float const , float*, unsigned short const *, int, int, int)
                    0.04%  40.021ms      5520  7.2500us  4.5760us  24.193us  block_scores_multi_kernel(float const *, float const *, float const *, int const *, int, long, float*, bool)
                    0.04%  39.737ms     22080  1.7990us  1.6640us  2.0800us  rebase_ptrs_kernel(__int64*, int const *, __int64, __int64)
                    0.04%  39.394ms       865  45.542us  42.210us  59.394us  gu_grouped_t_kernel(__int64 const *, int const *, int const *, int const *, unsigned char const *, float const *, float*, int)
                    0.04%  36.952ms        81  456.20us  149.06us  636.09us  void mul_mat_q<ggml_type=18, int=48, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.04%  34.262ms       820  41.783us  10.368us  61.059us  void native_mmvq_multi_kernel<SmallTraits<Q80Block, int=8>, int=4, int=4, int=2, bool=0, bool=0>(Q80BlockBlock const *, Q81Block const *, float*, int, int, Q80BlockBlock const , Q81Block const *)
                    0.03%  32.035ms      3060  10.468us  9.8890us  11.072us  void gdn_ab_multi_kernel<int=3, bool=1>(float const *, unsigned short const *, unsigned short const , float const , float const , float*, unsigned short const *, int, int, int)
                    0.03%  31.333ms       282  111.11us  103.43us  124.29us  void dequant_kernel<int=20, H16>(unsigned char const *, long, long, long, long, long, H16*)
                    0.03%  30.770ms     10327  2.9790us  1.9530us  19.648us  strata::kernels::copy_from_mapped_kernel(float4*, float4 volatile *, long)
                    0.03%  29.655ms       210  141.22us  20.289us  303.82us  void native_mmvq_multi_kernel<Q5KTraits, int=6, int=4, int=2, bool=0, bool=0>(Q5KTraitsBlock const *, Q81Block const *, float*, int, int, Q5KTraitsBlock const , Q81Block const *)
                    0.03%  26.401ms      2016  13.095us  12.417us  13.921us  void gdn_ab_multi_kernel<int=4, bool=1>(float const *, unsigned short const *, unsigned short const , float const , float const , float*, unsigned short const *, int, int, int)
                    0.03%  25.124ms       288  87.235us  14.337us  162.09us  swiglu_pair_kernel(float const *, float const *, unsigned short*, long)
                    0.03%  24.584ms       865  28.421us  25.761us  44.866us  down_grouped_t_kernel(__int64 const *, int const *, int const *, int const *, unsigned char const *, float const *, float*)
                    0.03%  24.430ms       109  224.13us  24.961us  287.24us  void mul_mat_q<ggml_type=20, int=40, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.02%  22.804ms        62  367.81us  94.532us  481.81us  void mul_mat_q<ggml_type=18, int=40, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.02%  22.443ms       261  85.988us  71.651us  96.420us  void native_mmvq_multi_kernel<SmallTraits<Q80Block, int=8>, int=8, int=4, int=2, bool=0, bool=0>(Q80BlockBlock const *, Q81Block const *, float*, int, int, Q80BlockBlock const , Q81Block const *)
                    0.02%  22.019ms       282  78.081us  19.489us  299.98us  void native_mmvq_multi_kernel<Q4KTraits, int=6, int=4, int=2, bool=0, bool=0>(Q4KTraitsBlock const *, Q81Block const *, float*, int, int, Q4KTraitsBlock const , Q81Block const *)
                    0.02%  21.086ms        18  1.1714ms  171.21us  2.2095ms  rms_rep_kernel(float const *, float const *, float*)
                    0.02%  20.825ms      6385  3.2610us  2.6240us  11.136us  gate(float const *, float const *, float*, int, int)
                    0.02%  20.374ms       252  80.848us  20.161us  299.08us  void native_mmvq_multi_kernel<IQ4XSTraits, int=6, int=4, int=2, bool=0, bool=0>(IQ4XSTraitsBlock const *, Q81Block const *, float*, int, int, IQ4XSTraitsBlock const , Q81Block const *)
                    0.02%  19.788ms      1346  14.701us  4.9600us  1.5624ms  embed_rows_kernel(int, unsigned char const *, unsigned long, int const *, long, float*)
                    0.02%  19.197ms       288  66.655us  20.161us  131.01us  void route_kernel<int=16>(float const *, int*, float*, long)
                    0.02%  18.390ms       225  81.734us  5.6000us  416.72us  void rope_kernel<bool=0>(float*, long, long, long, long, float, float, float, float, float, float, int const *, strata::kernels::RopeTab)
                    0.02%  16.898ms      1152  14.668us  2.9440us  59.331us  copy_i32_kernel(int*, int const *, long)
                    0.02%  16.282ms       732  22.242us  6.1760us  386.35us  void bf16_f32_mmvf_multi_kernel<int=256, int=6, bool=1>(float const *, long, unsigned short const *, float*, long, int, int)
                    0.02%  15.597ms       585  26.661us  5.2160us  53.922us  gr_silu_kernel(float const *, unsigned short*, unsigned short*, long)
                    0.02%  15.079ms        38  396.83us  146.79us  483.57us  void mul_mat_q<ggml_type=22, int=40, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.02%  14.533ms      1043  13.933us  12.288us  22.529us  void native_small_mmvq_kernel<Q80Block, int=8, bool=1>(Q80Block const *, Q81Block const *, float*, int, int)
                    0.01%  13.575ms         6  2.2626ms  343.57us  4.1859ms  conv_residual_batch_kernel(float const *, float const *, unsigned short const *, float*, float const *, int)
                    0.01%  13.447ms        28  480.26us  177.00us  627.45us  void mul_mat_q<ggml_type=21, int=48, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.01%  13.086ms       570  22.958us  8.8320us  48.930us  void native_mmvq_multi_kernel<SmallTraits<Q80Block, int=8>, int=2, int=4, int=2, bool=0, bool=0>(Q80BlockBlock const *, Q81Block const *, float*, int, int, Q80BlockBlock const , Q81Block const *)
                    0.01%  12.675ms      1742  7.2760us  4.4160us  21.313us  void norm<int=1024>(float const *, float const , float*, int, float)
                    0.01%  11.833ms        31  381.71us  57.091us  486.87us  void mul_mat_q<ggml_type=21, int=40, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.01%  11.316ms       880  12.858us  2.8480us  1.9722ms  add_streams_broadcast_kernel(float const *, float const *, float*, long, int)
                    0.01%  11.205ms      2889  3.8780us  3.2960us  5.1520us  void weighted_rms_norm<int=1024>(float const *, float const , float*, int, float)
                    0.01%  9.5658ms       282  33.921us  33.185us  35.105us  void native_mmvq_multi_kernel<SmallTraits<IQ4NLBlock, int=4>, int=6, int=4, int=2, bool=0, bool=0>(IQ4NLBlockBlock const *, Q81Block const *, float*, int, int, IQ4NLBlockBlock const , Q81Block const *)
                    0.01%  9.2924ms       865  10.742us  9.7290us  23.169us  mtp_select_kernel(float const *, long, int const *, int const *, float*, int*, int*, int, float const *, float*)
                    0.01%  9.2746ms        35  264.99us  49.826us  345.17us  void mul_mat_q<ggml_type=42, int=48, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.01%  8.8519ms       969  9.1350us  2.3040us  1.8318ms  gate_kernel(float const *, float const *, float*, float)
                    0.01%  8.8496ms       963  9.1890us  8.7360us  10.017us  history_advance_kernel(float*, float const *)
                    0.01%  7.9552ms       865  9.1960us  7.6480us  23.617us  group_resident_kernel(int const *, int, int, unsigned char const *, __int64, __int64*, int*, int*, int*, int*)
                    0.01%  7.6550ms       865  8.8490us  8.6080us  20.896us  row_top_prob_split_kernel(float const *, int, int const *, float*, TopRow*)
                    0.01%  7.4022ms        30  246.74us  76.804us  366.51us  void mul_mat_q<ggml_type=18, int=32, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.01%  6.8682ms       211  32.550us  1.7920us  1.6933ms  [CUDA memcpy DtoD]
                    0.01%  6.8484ms       360  19.023us  16.737us  23.521us  void cublasLt::splitKreduce_kernel<int=32, int=16, int, float, float, float, float, bool=1, bool=0, bool=0>(cublasLt::cublasSplitKParams<float>, float const *, float const *, float const **, cublasLt::cublasSplitKParams const *, cublasLt::cublasSplitKParams const , float const *, float const , cublasLt::cublasSplitKParams const **, void*, long, cublasLt::cublasSplitKParams*, int*)
                    0.01%  6.2034ms      1785  3.4750us  3.1360us  10.464us  post(float const *, float const *, float const *, float*, int, int, float)
                    0.01%  5.9702ms         6  995.04us  148.04us  1.8592ms  broadcast_batch_kernel(float const *, float const *, float*, int)
                    0.01%  5.5813ms       247  22.596us  9.4720us  51.938us  void native_mmvq_multi_kernel<SmallTraits<Q80Block, int=8>, int=3, int=4, int=2, bool=0, bool=0>(Q80BlockBlock const *, Q81Block const *, float*, int, int, Q80BlockBlock const , Q81Block const *)
                    0.01%  5.3085ms         6  884.76us  132.97us  1.6393ms  gr_broadcast_kernel(float const *, float*, long)
                    0.01%  5.0026ms        24  208.44us  100.33us  266.41us  void mul_mat_q<ggml_type=42, int=40, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.00%  4.5550ms        34  133.97us  32.033us  198.02us  void mul_mat_q<ggml_type=20, int=32, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.00%  4.4299ms       865  5.1210us  3.2640us  17.057us  strata::kernels::copy_from_mapped_multi_kernel(strata::kernels::MappedCopies)
                    0.00%  4.2614ms       216  19.728us  19.361us  20.193us  void gdn_ab_multi_kernel<int=6, bool=1>(float const *, unsigned short const *, unsigned short const , float const , float const , float*, unsigned short const *, int, int, int)
                    0.00%  3.9744ms       865  4.5940us  3.7440us  10.785us  swiglu_quantize_q8_0_scaled_kernel(float*, __int64, unsigned char*, float*)
                    0.00%  3.7185ms       460  8.0830us  6.8160us  20.641us  logits_nonfinite_kernel(float const *, int, int, unsigned int*)
                    0.00%  3.5315ms       963  3.6670us  3.1680us  4.7040us  conv_residual_kernel(float const *, float const *, unsigned short const *, float const *, float const *, float*, float*)
                    0.00%  3.2981ms       865  3.8120us  2.7840us  10.976us  quantize_q8_0_scaled_kernel(float const *, unsigned char*, float*, __int64)
                    0.00%  3.0594ms        48  63.737us  38.689us  80.067us  void gr_down_multi_kernel<int=2560, int=8>(GrMulti)
                    0.00%  2.8767ms         9  319.64us  128.36us  359.34us  void mul_mat_q<ggml_type=22, int=32, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.00%  2.8028ms       221  12.682us  5.4720us  21.505us  copy_row_to_first_kernel(int const *, float*, long, float*, long, float*, long)
                    0.00%  2.5620ms       576  4.4470us  1.9200us  11.201us  void apply<bool=0>(float const *, float*, int, int, int, float, float, float, float, float, float, int const *, int const , strata::kernels::RopeTab)
                    0.00%  2.4951ms       865  2.8840us  2.2400us  8.1280us  window_ids_kernel(int*, int, int*, __int64)
                    0.00%  2.4254ms       216  11.228us  3.6800us  21.569us  gdn_gates_kernel(float const *, float const *, float const *, float*, float*, long)
                    0.00%  2.2568ms       282  8.0020us  5.4400us  23.297us  copy_indexed_kernel(float*, float const *, long, int const *, long)
                    0.00%  2.0714ms         1  2.0714ms  2.0714ms  2.0714ms  void gather_rows_kernel<unsigned int>(unsigned int const *, __int64, int const *, __int64, gather_rows_kernel<unsigned int>*)
                    0.00%  2.0578ms       963  2.1360us  1.8880us  2.4010us  broadcast_kernel(float const *, float const *, float*)
                    0.00%  1.8639ms       865  2.1540us  2.0480us  8.9280us  map_ids_kernel(int*, int const *, int)
                    0.00%  1.5668ms         8  195.85us  74.851us  275.60us  void mul_mat_q<ggml_type=18, int=24, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.00%  1.3734ms       227  6.0500us  4.0640us  14.144us  void norm<int=256>(float const *, float const , float*, int, float)
                    0.00%  1.2722ms         3  424.08us  393.07us  480.98us  void mul_mat_q<ggml_type=23, int=48, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.00%  1.2578ms       460  2.7340us  2.4960us  3.3280us  broadcast_streams_kernel(float const *, float*, long, int)
                    0.00%  1.2074ms         8  150.93us  96.132us  168.30us  void mul_mat_q<ggml_type=42, int=32, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.00%  1.2001ms         6  200.01us  35.362us  364.53us  to_bf16_kernel(float const *, unsigned short*, unsigned short*, long)
                    0.00%  1.1690ms        72  16.236us  3.5840us  32.162us  void append_blocks<bool=0>(float const *, long, long, int, float const , float, float const , float const , float*, int*, long, long, float, float, float, float, float, float, int const *, strata::kernels::RopeTab)
                    0.00%  1.0715ms        12  89.291us  27.969us  149.00us  void mul_mat_q<ggml_type=20, int=24, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.00%  936.74us        12  78.062us  71.235us  84.548us  void gr_up_multi_kernel<int=8, bool=0>(GrMulti)
                    0.00%  891.56us        24  37.148us  22.849us  68.451us  void native_mmvq_multi_kernel<SmallTraits<Q80Block, int=8>, int=6, int=4, int=2, bool=0, bool=0>(Q80BlockBlock const *, Q81Block const *, float*, int, int, Q80BlockBlock const , Q81Block const *)
                    0.00%  859.56us         5  171.91us  62.307us  272.56us  void mul_mat_q<ggml_type=22, int=24, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.00%  802.24us         3  267.41us  156.42us  323.34us  void mul_mat_q<ggml_type=21, int=32, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.00%  794.62us       216  3.6780us  3.2640us  13.984us  gdn_conv_hist_kernel(float*, float const *, long)
                    0.00%  710.91us         5  142.18us  78.435us  236.36us  void mul_mat_q<ggml_type=22, int=16, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.00%  671.04us         2  335.52us  309.39us  361.65us  void mul_mat_q<ggml_type=23, int=40, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.00%  476.82us         6  79.470us  70.595us  88.068us  void gr_up_multi_kernel<int=5, bool=1>(GrMulti)
                    0.00%  359.05us        13  27.619us  18.432us  47.554us  void mul_mat_q<ggml_type=20, int=8, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.00%  349.30us         7  49.899us  34.370us  91.396us  void mul_mat_q<ggml_type=22, int=8, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.00%  347.79us         5  69.558us  41.826us  131.17us  void mul_mat_q<ggml_type=20, int=16, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.00%  336.27us         6  56.045us  37.058us  95.268us  void mul_mat_q<ggml_type=18, int=8, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.00%  260.24us         2  130.12us  113.51us  146.73us  void mul_mat_q<ggml_type=42, int=24, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.00%  240.97us        96  2.5100us  2.2400us  5.2800us  reset_kernel(strata::kernels::KvStreamMap)
                    0.00%  231.14us        16  14.446us  12.320us  16.896us  gr_norm_multi_kernel(GrMulti)
                    0.00%  227.95us         3  75.982us  63.235us  90.244us  void mul_mat_q<ggml_type=18, int=16, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.00%  196.62us        72  2.7300us  2.3680us  3.8720us  void append_first<bool=0>(float const *, float const , float, float*, float const *, float, float, float, float, float, float, int const *, strata::kernels::RopeTab)
                    0.00%  168.26us         3  56.088us  45.570us  72.643us  void mul_mat_q<ggml_type=42, int=16, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.00%  151.49us         1  151.49us  151.49us  151.49us  void mul_mat_q<ggml_type=21, int=24, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.00%  138.60us        72  1.9240us  1.5360us  2.7520us  append_tail(float const *, long, long, float*)
                    0.00%  104.32us         6  17.387us  16.609us  18.209us  void dequant_q8_0_kernel<H16>(unsigned char const *, long, long, long, long, H16*)
                    0.00%  102.73us         2  51.362us  40.930us  61.795us  void mul_mat_q<ggml_type=21, int=8, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.00%  71.330us         2  35.665us  34.209us  37.121us  gr_down_kernel(strata::kernels::FusedGrArgs)
                    0.00%  68.996us         2  34.498us  31.682us  37.314us  gr_up_kernel(strata::kernels::FusedGrArgs)
                    0.00%  60.965us         6  10.160us  10.016us  10.305us  history_batch_kernel(float*, float const *, int)
                    0.00%  59.971us         2  29.985us  18.881us  41.090us  void mul_mat_q<ggml_type=42, int=8, bool=0>(char const *, int const *, int const *, int const *, float*, float*, float const *, uint3, int, int, int, int, int, uint3, uint3, int, int, int, uint3, uint3, int, int, int, uint3)
                    0.00%  22.304us        12  1.8580us  1.5680us  2.2400us  iota_kernel(int*, long)
                    0.00%  5.7280us         1  5.7280us  5.7280us  5.7280us  ring_kernel(int*, __int64, __int64)
      API calls:   72.73%  57.5789s      2311  24.915ms     925ns  486.64ms  cudaStreamSynchronize
                   12.65%  10.0172s         1  10.0172s  10.0172s  10.0172s  cudaHostRegister
                    3.25%  2.57530s      1613  1.5966ms  80.307us  12.039ms  cudaGraphLaunch
                    2.77%  2.19627s      9519  230.72us  6.3160us  92.873ms  cudaMemcpy
                    2.21%  1.75303s       102  17.187ms  3.6050us  333.57ms  cudaHostAlloc
                    1.76%  1.39535s         1  1.39535s  1.39535s  1.39535s  cudaHostUnregister
                    1.31%  1.03335s     75756  13.640us  2.1320us  11.684ms  cudaMemcpyAsync
                    0.79%  627.04ms    111034  5.6470us  1.0990us  5.1993ms  cudaLaunchKernel
                    0.67%  532.87ms       840  634.37us  2.3280us  16.214ms  cuLibraryLoadData
                    0.44%  349.49ms       152  2.2993ms  1.0720us  14.530ms  cudaEventSynchronize
                    0.41%  322.96ms       325  993.72us  3.3320us  105.07ms  cudaFree
                    0.17%  131.25ms    133532     982ns     577ns  932.01us  cudaStreamQuery
                    0.14%  112.54ms    109712  1.0250us     487ns  6.1790ms  cudaEventRecord
                    0.14%  111.56ms        33  3.3807ms  1.2690us  106.07ms  cudaDeviceSynchronize
                    0.11%  90.434ms        58  1.5592ms  2.6850us  41.455ms  cudaFreeHost
                    0.10%  80.396ms     85393     941ns     546ns  5.0005ms  cudaStreamWaitEvent
                    0.10%  77.039ms        13  5.9261ms  135.21us  18.463ms  cudaGraphInstantiate
                    0.06%  50.793ms       345  147.23us  5.0700us  7.1931ms  cudaMalloc
                    0.05%  39.806ms     11045  3.6030us  1.1590us  4.0262ms  cudaMemsetAsync
                    0.03%  26.792ms     56868     471ns     190ns  1.9371ms  cudaGetDevice
                    0.02%  18.994ms        13  1.4611ms  56.948us  3.5016ms  cudaGraphExecDestroy
                    0.02%  13.623ms     90989     149ns      88ns  914.77us  cudaGetLastError
                    0.01%  11.292ms        12  941.02us  36.599us  2.6578ms  cudaGraphUpload
                    0.01%  7.2248ms         1  7.2248ms  7.2248ms  7.2248ms  cudaGetSymbolAddress
                    0.01%  5.1335ms        13  394.88us  13.327us  1.5816ms  cudaGraphDestroy
                    0.00%  1.2095ms        13  93.040us  3.5320us  263.02us  cudaStreamEndCapture
                    0.00%  1.1356ms       276  4.1140us  1.0860us  32.397us  cudaEventElapsedTime
                    0.00%  944.11us        18  52.450us  10.075us  94.039us  cudaMemGetInfo
                    0.00%  714.45us      1082     660ns     316ns  18.403us  cudaEventDestroy
                    0.00%  692.30us      1076     643ns     396ns  18.055us  cudaEventCreateWithFlags
                    0.00%  465.17us         3  155.06us  127.32us  196.18us  cudaGetDeviceProperties
                    0.00%  426.85us       336  1.2700us     124ns  55.613us  cuDeviceGetAttribute
                    0.00%  374.09us       294  1.2720us     350ns  17.693us  cudaOccupancyMaxActiveBlocksPerMultiprocessor
                    0.00%  366.46us        10  36.646us  2.6200us  241.22us  cudaStreamCreateWithFlags
                    0.00%  259.89us         1  259.89us  259.89us  259.89us  cudaLaunchHostFunc
                    0.00%  239.12us       810     295ns     128ns  17.956us  cuGetProcAddress
                    0.00%  200.88us        85  2.3630us     836ns  20.574us  cudaFuncSetAttribute
                    0.00%  139.59us         5  27.918us  8.6910us  60.795us  cudaMemset
                    0.00%  114.41us        44  2.6000us     404ns  16.080us  cudaHostGetDevicePointer
                    0.00%  101.48us         4  25.369us  16.955us  34.216us  cuDeviceGetName
                    0.00%  86.476us       451     191ns     107ns     489ns  cudaPeekAtLastError
                    0.00%  64.638us        13  4.9720us  1.7100us  8.6090us  cudaStreamBeginCapture
                    0.00%  55.793us         7  7.9700us  2.8940us  15.221us  cudaStreamDestroy
                    0.00%  47.916us        50     958ns     209ns  17.301us  cudaDeviceGetAttribute
                    0.00%  43.809us         7  6.2580us  1.2000us  19.487us  cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags
                    0.00%  22.746us         5  4.5490us  1.4480us  12.216us  cudaSetDevice
                    0.00%  18.421us         8  2.3020us     745ns  6.4930us  cudaEventCreate
                    0.00%  14.124us         1  14.124us  14.124us  14.124us  cudaFuncGetAttributes
                    0.00%  8.9660us         3  2.9880us  2.8230us  3.2960us  cuInit
                    0.00%  6.2470us         1  6.2470us  6.2470us  6.2470us  cuDeviceGetPCIBusId
                    0.00%  3.8420us        10     384ns      80ns  1.0980us  cudaGetErrorString
                    0.00%  2.2600us         2  1.1300us     558ns  1.7020us  cudaGetDriverEntryPoint
                    0.00%  1.8180us         3     606ns     517ns     707ns  cuDeviceTotalMem
                    0.00%  1.7590us         4     439ns     224ns     906ns  cuDeviceGetCount
                    0.00%  1.2580us         3     419ns     249ns     590ns  cuModuleGetLoadingMode
                    0.00%     897ns         5     179ns     143ns     232ns  cuDeviceGet
                    0.00%     726ns         3     242ns     215ns     269ns  cuDeviceGetUuid
                    0.00%     421ns         1     421ns     421ns     421ns  cudaGetDeviceCount
                    0.00%     381ns         2     190ns     184ns     197ns  cuDriverGetVersion
                    0.00%     333ns         1     333ns     333ns     333ns  cudaRuntimeGetVersion
