Port Qwen row-map SwiGLU expert gathers

This commit is contained in:
Georg Bauer committed 2026-10-04 16:24:33 +02:00
1 parent 8965e77ea1
commit 7b894fbfee
42 files changed
+11489 -65

No files matched your search

@@ -0,0 +1,838 @@
[
{
"configuration": "default-tail",
"arm": "before",
"records": [
{
"chunk": 2048,
"rep": 0,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 3.7007944999999998,
"prefill_seconds": 3.6969075010000005,
"prefill_tokens_per_second": 1107.9530658779117,
"elapsed_seconds": 3.721188
},
{
"chunk": 2048,
"rep": 0,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 1.7724733750000001,
"prefill_seconds": 1.769735542,
"prefill_tokens_per_second": 1157.23505088536,
"elapsed_seconds": 1.793335125
},
{
"chunk": 2048,
"rep": 1,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 2.820473209,
"prefill_seconds": 2.81689246,
"prefill_tokens_per_second": 1454.0846191906098,
"elapsed_seconds": 2.841268834
},
{
"chunk": 2048,
"rep": 1,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 1.838782375,
"prefill_seconds": 1.836070168,
"prefill_tokens_per_second": 1115.4257803942492,
"elapsed_seconds": 1.8597940419999999
},
{
"chunk": 2048,
"rep": 2,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 2.959423792,
"prefill_seconds": 2.9556270000000002,
"prefill_tokens_per_second": 1385.831162051233,
"elapsed_seconds": 2.98031925
},
{
"chunk": 2048,
"rep": 2,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 1.922527375,
"prefill_seconds": 1.919643709,
"prefill_tokens_per_second": 1066.8646428492007,
"elapsed_seconds": 1.9441567499999999
},
{
"chunk": 4096,
"rep": 0,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 3.626180458,
"prefill_seconds": 3.6201352510000007,
"prefill_tokens_per_second": 1131.4494393181994,
"elapsed_seconds": 3.64758525
},
{
"chunk": 4096,
"rep": 0,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 1.994229042,
"prefill_seconds": 1.9914031639999998,
"prefill_tokens_per_second": 1028.4205815392588,
"elapsed_seconds": 2.016047875
},
{
"chunk": 4096,
"rep": 1,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 3.6903119589999998,
"prefill_seconds": 3.6843170420000004,
"prefill_tokens_per_second": 1111.7392866322168,
"elapsed_seconds": 3.711886
},
{
"chunk": 4096,
"rep": 1,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 2.011648875,
"prefill_seconds": 2.0087712079999998,
"prefill_tokens_per_second": 1019.5287506331085,
"elapsed_seconds": 2.033881333
},
{
"chunk": 4096,
"rep": 2,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 3.752376958,
"prefill_seconds": 3.7459722089999996,
"prefill_tokens_per_second": 1093.4411072669013,
"elapsed_seconds": 3.774499291
},
{
"chunk": 4096,
"rep": 2,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 2.020803542,
"prefill_seconds": 2.017895623,
"prefill_tokens_per_second": 1014.918698795354,
"elapsed_seconds": 2.043202292
}
],
"summary": [
{
"chunk": 2048,
"phase": "cold",
"ttft_seconds": {
"mean": 3.1602305003333333,
"median": 2.959423792,
"sample_stddev": 0.47326936772912714
},
"prefill_tokens_per_second": {
"mean": 1315.9562823732515,
"median": 1385.831162051233,
"sample_stddev": 183.340222420435
}
},
{
"chunk": 2048,
"phase": "warm",
"ttft_seconds": {
"mean": 1.844594375,
"median": 1.838782375,
"sample_stddev": 0.07519564639658331
},
"prefill_tokens_per_second": {
"mean": 1113.1751580429366,
"median": 1115.4257803942492,
"sample_stddev": 45.227222310038556
}
},
{
"chunk": 4096,
"phase": "cold",
"ttft_seconds": {
"mean": 3.689623125,
"median": 3.6903119589999998,
"sample_stddev": 0.06310106989799923
},
"prefill_tokens_per_second": {
"mean": 1112.2099444057726,
"median": 1111.7392866322168,
"sample_stddev": 19.00853664502584
}
},
{
"chunk": 4096,
"phase": "warm",
"ttft_seconds": {
"mean": 2.0088938196666666,
"median": 2.011648875,
"sample_stddev": 0.013499768886162316
},
"prefill_tokens_per_second": {
"mean": 1020.9560103225738,
"median": 1019.5287506331085,
"sample_stddev": 6.863163415905101
}
}
],
"resources": {
"event": "test_resource_summary",
"elapsed_ms": 41097,
"peak": {
"cpu_time_ns": 19599062416,
"resident_bytes": 25489604608,
"physical_bytes": 84724073856,
"peak_physical_bytes": 84724139512,
"disk_read_bytes": 64097386496,
"disk_write_bytes": 442368
},
"error": null
}
},
{
"configuration": "default-tail",
"arm": "after",
"records": [
{
"chunk": 2048,
"rep": 0,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 3.699041667,
"prefill_seconds": 3.6953074989999997,
"prefill_tokens_per_second": 1108.4327897227588,
"elapsed_seconds": 3.719328292
},
{
"chunk": 2048,
"rep": 0,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 1.776398,
"prefill_seconds": 1.7737045,
"prefill_tokens_per_second": 1154.6455455235075,
"elapsed_seconds": 1.797202167
},
{
"chunk": 2048,
"rep": 1,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 2.80434375,
"prefill_seconds": 2.8008109150000005,
"prefill_tokens_per_second": 1462.4336038050606,
"elapsed_seconds": 2.825135334
},
{
"chunk": 2048,
"rep": 1,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 1.900156542,
"prefill_seconds": 1.897415875,
"prefill_tokens_per_second": 1079.3627411808177,
"elapsed_seconds": 1.9216478750000001
},
{
"chunk": 2048,
"rep": 2,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 2.952652333,
"prefill_seconds": 2.948979292,
"prefill_tokens_per_second": 1388.9551585226932,
"elapsed_seconds": 2.973752041
},
{
"chunk": 2048,
"rep": 2,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 1.947212708,
"prefill_seconds": 1.944330169,
"prefill_tokens_per_second": 1053.3190466582735,
"elapsed_seconds": 1.9688718330000001
},
{
"chunk": 4096,
"rep": 0,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 3.637618958,
"prefill_seconds": 3.631528458,
"prefill_tokens_per_second": 1127.8997390139687,
"elapsed_seconds": 3.659774833
},
{
"chunk": 4096,
"rep": 0,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 1.9611875410000001,
"prefill_seconds": 1.958409373,
"prefill_tokens_per_second": 1045.7466289914453,
"elapsed_seconds": 1.982932208
},
{
"chunk": 4096,
"rep": 1,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 3.6904271250000003,
"prefill_seconds": 3.684310291,
"prefill_tokens_per_second": 1111.7413237440048,
"elapsed_seconds": 3.71194925
},
{
"chunk": 4096,
"rep": 1,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 1.988992875,
"prefill_seconds": 1.98611425,
"prefill_tokens_per_second": 1031.159209496634,
"elapsed_seconds": 2.010877
},
{
"chunk": 4096,
"rep": 2,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 3.684088083,
"prefill_seconds": 3.6781829150000003,
"prefill_tokens_per_second": 1113.5933406944227,
"elapsed_seconds": 3.705752166
},
{
"chunk": 4096,
"rep": 2,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 1.998740333,
"prefill_seconds": 1.9958163330000003,
"prefill_tokens_per_second": 1026.1465276825147,
"elapsed_seconds": 2.021001667
}
],
"summary": [
{
"chunk": 2048,
"phase": "cold",
"ttft_seconds": {
"mean": 3.1520125833333332,
"median": 2.952652333,
"sample_stddev": 0.4795096169325091
},
"prefill_tokens_per_second": {
"mean": 1319.940517350171,
"median": 1388.9551585226932,
"sample_stddev": 186.81918963228082
}
},
{
"chunk": 2048,
"phase": "warm",
"ttft_seconds": {
"mean": 1.8745890833333332,
"median": 1.900156542,
"sample_stddev": 0.0882308751191163
},
"prefill_tokens_per_second": {
"mean": 1095.775777787533,
"median": 1079.3627411808177,
"sample_stddev": 52.619441949355554
}
},
{
"chunk": 4096,
"phase": "cold",
"ttft_seconds": {
"mean": 3.6707113886666667,
"median": 3.684088083,
"sample_stddev": 0.02883361907462447
},
"prefill_tokens_per_second": {
"mean": 1117.7448011507988,
"median": 1113.5933406944227,
"sample_stddev": 8.843051732845977
}
},
{
"chunk": 4096,
"phase": "warm",
"ttft_seconds": {
"mean": 1.9829735830000002,
"median": 1.988992875,
"sample_stddev": 0.019486583946006573
},
"prefill_tokens_per_second": {
"mean": 1034.3507887235314,
"median": 1031.159209496634,
"sample_stddev": 10.18236840333609
}
}
],
"resources": {
"event": "test_resource_summary",
"elapsed_ms": 40871,
"peak": {
"cpu_time_ns": 19458778041,
"resident_bytes": 25328418816,
"physical_bytes": 84668646808,
"peak_physical_bytes": 84668761376,
"disk_read_bytes": 69973549056,
"disk_write_bytes": 18446744073709289472
},
"error": null
}
},
{
"configuration": "full-chunks",
"arm": "before",
"records": [
{
"chunk": 2048,
"rep": 0,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 3.182152,
"prefill_seconds": 3.17999279,
"prefill_tokens_per_second": 1288.0532348628376,
"elapsed_seconds": 3.202144875
},
{
"chunk": 2048,
"rep": 0,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 1.219195625,
"prefill_seconds": 1.21829175,
"prefill_tokens_per_second": 1681.042328325707,
"elapsed_seconds": 1.240087208
},
{
"chunk": 2048,
"rep": 1,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 2.317122125,
"prefill_seconds": 2.3159108330000002,
"prefill_tokens_per_second": 1768.6345871503593,
"elapsed_seconds": 2.338011208
},
{
"chunk": 2048,
"rep": 1,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 1.26216725,
"prefill_seconds": 1.261251084,
"prefill_tokens_per_second": 1623.7845310743853,
"elapsed_seconds": 1.283571459
},
{
"chunk": 2048,
"rep": 2,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 2.411353459,
"prefill_seconds": 2.410105791,
"prefill_tokens_per_second": 1699.510459373026,
"elapsed_seconds": 2.431755125
},
{
"chunk": 2048,
"rep": 2,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 1.34331275,
"prefill_seconds": 1.34238875,
"prefill_tokens_per_second": 1525.6385305672443,
"elapsed_seconds": 1.364779
},
{
"chunk": 4096,
"rep": 0,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 3.2133175,
"prefill_seconds": 3.212275166,
"prefill_tokens_per_second": 1275.1086965879188,
"elapsed_seconds": 3.233724875
},
{
"chunk": 4096,
"rep": 0,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 1.7273878329999999,
"prefill_seconds": 1.7264041659999998,
"prefill_tokens_per_second": 1186.2807332915115,
"elapsed_seconds": 1.7481919160000001
},
{
"chunk": 4096,
"rep": 1,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 2.980801042,
"prefill_seconds": 2.979894292,
"prefill_tokens_per_second": 1374.5454028340412,
"elapsed_seconds": 3.001205167
},
{
"chunk": 4096,
"rep": 1,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 1.7668409999999999,
"prefill_seconds": 1.765948334,
"prefill_tokens_per_second": 1159.7168278197203,
"elapsed_seconds": 1.787861709
},
{
"chunk": 4096,
"rep": 2,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 3.100441792,
"prefill_seconds": 3.0994622499999998,
"prefill_tokens_per_second": 1321.5195635952657,
"elapsed_seconds": 3.121023875
},
{
"chunk": 4096,
"rep": 2,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 1.838995291,
"prefill_seconds": 1.838062583,
"prefill_tokens_per_second": 1114.216686059378,
"elapsed_seconds": 1.860065208
}
],
"summary": [
{
"chunk": 2048,
"phase": "cold",
"ttft_seconds": {
"mean": 2.6368758613333334,
"median": 2.411353459,
"sample_stddev": 0.4745676312458291
},
"prefill_tokens_per_second": {
"mean": 1585.3994271287409,
"median": 1699.510459373026,
"sample_stddev": 259.81840735698063
}
},
{
"chunk": 2048,
"phase": "warm",
"ttft_seconds": {
"mean": 1.274891875,
"median": 1.26216725,
"sample_stddev": 0.06302937601316841
},
"prefill_tokens_per_second": {
"mean": 1610.1551299891123,
"median": 1623.7845310743853,
"sample_stddev": 78.59329182503896
}
},
{
"chunk": 4096,
"phase": "cold",
"ttft_seconds": {
"mean": 3.098186778,
"median": 3.100441792,
"sample_stddev": 0.11627463019215153
},
"prefill_tokens_per_second": {
"mean": 1323.7245543390752,
"median": 1321.5195635952657,
"sample_stddev": 49.755011058230345
}
},
{
"chunk": 4096,
"phase": "warm",
"ttft_seconds": {
"mean": 1.7777413746666666,
"median": 1.7668409999999999,
"sample_stddev": 0.05659655286508827
},
"prefill_tokens_per_second": {
"mean": 1153.4047490568698,
"median": 1159.7168278197203,
"sample_stddev": 36.444320265304924
}
}
],
"resources": {
"event": "test_resource_summary",
"elapsed_ms": 35856,
"peak": {
"cpu_time_ns": 21540742125,
"resident_bytes": 25113673728,
"physical_bytes": 88578784936,
"peak_physical_bytes": 88578866928,
"disk_read_bytes": 69525413888,
"disk_write_bytes": 524288
},
"error": null
}
},
{
"configuration": "full-chunks",
"arm": "after",
"records": [
{
"chunk": 2048,
"rep": 0,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 3.126132542,
"prefill_seconds": 3.1225067920000003,
"prefill_tokens_per_second": 1311.766562203846,
"elapsed_seconds": 3.147092209
},
{
"chunk": 2048,
"rep": 0,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 1.262224125,
"prefill_seconds": 1.258582084,
"prefill_tokens_per_second": 1627.2279941337542,
"elapsed_seconds": 1.282883709
},
{
"chunk": 2048,
"rep": 1,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 2.304937333,
"prefill_seconds": 2.303636208,
"prefill_tokens_per_second": 1778.058525810426,
"elapsed_seconds": 2.325319833
},
{
"chunk": 2048,
"rep": 1,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 1.333947917,
"prefill_seconds": 1.3330325,
"prefill_tokens_per_second": 1536.346638210246,
"elapsed_seconds": 1.354474625
},
{
"chunk": 2048,
"rep": 2,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 2.504060958,
"prefill_seconds": 2.502749457,
"prefill_tokens_per_second": 1636.6000953646394,
"elapsed_seconds": 2.524487958
},
{
"chunk": 2048,
"rep": 2,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 1.363442375,
"prefill_seconds": 1.362460209,
"prefill_tokens_per_second": 1503.1631650388993,
"elapsed_seconds": 1.3840412500000001
},
{
"chunk": 4096,
"rep": 0,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 2.955218083,
"prefill_seconds": 2.9541866239999997,
"prefill_tokens_per_second": 1386.50685326507,
"elapsed_seconds": 2.9757388750000002
},
{
"chunk": 4096,
"rep": 0,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 1.606924,
"prefill_seconds": 1.6059842079999997,
"prefill_tokens_per_second": 1275.2304722538095,
"elapsed_seconds": 1.62792175
},
{
"chunk": 4096,
"rep": 1,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 3.209116167,
"prefill_seconds": 3.2081535000000003,
"prefill_tokens_per_second": 1276.7468888256126,
"elapsed_seconds": 3.230131833
},
{
"chunk": 4096,
"rep": 1,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 1.767990417,
"prefill_seconds": 1.767019791,
"prefill_tokens_per_second": 1159.0136174088839,
"elapsed_seconds": 1.789304875
},
{
"chunk": 4096,
"rep": 2,
"phase": "cold",
"prompt": 4096,
"cached": 0,
"ttft_seconds": 2.998665042,
"prefill_seconds": 2.997745792,
"prefill_tokens_per_second": 1366.3600198959098,
"elapsed_seconds": 3.019893375
},
{
"chunk": 4096,
"rep": 2,
"phase": "warm",
"prompt": 6145,
"cached": 4097,
"ttft_seconds": 1.633976833,
"prefill_seconds": 1.633039209,
"prefill_tokens_per_second": 1254.1033850951462,
"elapsed_seconds": 1.656772416
}
],
"summary": [
{
"chunk": 2048,
"phase": "cold",
"ttft_seconds": {
"mean": 2.645043611,
"median": 2.504060958,
"sample_stddev": 0.4283660515917751
},
"prefill_tokens_per_second": {
"mean": 1575.475061126304,
"median": 1636.6000953646394,
"sample_stddev": 239.08001001547228
}
},
{
"chunk": 2048,
"phase": "warm",
"ttft_seconds": {
"mean": 1.3198714723333334,
"median": 1.333947917,
"sample_stddev": 0.052056635063226324
},
"prefill_tokens_per_second": {
"mean": 1555.5792657942998,
"median": 1536.346638210246,
"sample_stddev": 64.22959541694213
}
},
{
"chunk": 4096,
"phase": "cold",
"ttft_seconds": {
"mean": 3.054333097333333,
"median": 2.998665042,
"sample_stddev": 0.1357949135761868
},
"prefill_tokens_per_second": {
"mean": 1343.2045873288641,
"median": 1366.3600198959098,
"sample_stddev": 58.42895678910406
}
},
{
"chunk": 4096,
"phase": "warm",
"ttft_seconds": {
"mean": 1.6696304166666667,
"median": 1.633976833,
"sample_stddev": 0.0862495286509525
},
"prefill_tokens_per_second": {
"mean": 1229.4491582526132,
"median": 1254.1033850951462,
"sample_stddev": 61.90688583198103
}
}
],
"resources": {
"event": "test_resource_summary",
"elapsed_ms": 34614,
"peak": {
"cpu_time_ns": 20995139625,
"resident_bytes": 25576030208,
"physical_bytes": 87952375368,
"peak_physical_bytes": 87952457288,
"disk_read_bytes": 64869085184,
"disk_write_bytes": 442368
},
"error": null
}
}
]