diff --git a/.github/prediction-numerical-sentinels.json b/.github/prediction-numerical-sentinels.json index 421b3074c3..4230d8692a 100644 --- a/.github/prediction-numerical-sentinels.json +++ b/.github/prediction-numerical-sentinels.json @@ -1,6 +1,6 @@ { "schema_version": 1, - "baseline_source_sha": "a4c77310d7609745863ac8867c207f850beac5c3", + "baseline_source_sha": "af91885dfdeddedebd256c0c4df11a6f3e75e97e", "interpretation": "Native prediction stability sentinels; not measured-hardware accuracy. Two percent relative and 0.0001 ms absolute tolerance permit small numerical changes; baseline changes require explained review.", "cases": [ { @@ -102,7 +102,7 @@ "moe_tp_size": 1, "moe_ep_size": 8, "database_mode": "SILICON", - "shared_layer": false, + "shared_layer": true, "forward_model": "op_level" }, "method": "predict_prefill_latency", @@ -111,7 +111,7 @@ "isl": 1024, "prefix": 0 }, - "expected_ms": 42.67633133937119, + "expected_ms": 21.898890264920386, "rtol": 0.02, "atol_ms": 0.0001 }, @@ -126,7 +126,7 @@ "moe_tp_size": 1, "moe_ep_size": 8, "database_mode": "SILICON", - "shared_layer": false, + "shared_layer": true, "forward_model": "op_level" }, "method": "predict_prefill_latency", @@ -135,7 +135,7 @@ "isl": 8192, "prefix": 128 }, - "expected_ms": 116.4093545460498, + "expected_ms": 98.96825693463862, "rtol": 0.02, "atol_ms": 0.0001 }, @@ -150,7 +150,7 @@ "moe_tp_size": 1, "moe_ep_size": 8, "database_mode": "SILICON", - "shared_layer": false, + "shared_layer": true, "forward_model": "op_level" }, "method": "predict_decode_latency", @@ -159,7 +159,7 @@ "isl": 1024, "osl": 2 }, - "expected_ms": 39.36455865490505, + "expected_ms": 6.956580007206089, "rtol": 0.02, "atol_ms": 0.0001 }, @@ -174,7 +174,7 @@ "moe_tp_size": 1, "moe_ep_size": 8, "database_mode": "SILICON", - "shared_layer": false, + "shared_layer": true, "forward_model": "op_level" }, "method": "predict_decode_latency", @@ -183,7 +183,7 @@ "isl": 8192, "osl": 256 }, - "expected_ms": 7396.437641058326, + "expected_ms": 2192.9660386705996, "rtol": 0.02, "atol_ms": 0.0001 } diff --git a/crates/core/parity_tests/perfmodel/README.md b/crates/core/parity_tests/perfmodel/README.md index 2dae18830a..d213ea9182 100644 --- a/crates/core/parity_tests/perfmodel/README.md +++ b/crates/core/parity_tests/perfmodel/README.md @@ -97,6 +97,26 @@ requirement, pinned thread caps, byte-reproducible output, and all-payloads-before-any-write. `TestGoldenComparisonGuards` proves the comparison itself still bites. +### FP8-block correction in PR #244 + +The selective refresh from `0a51476b6ab90bc0e475bd41d4b1c7abbef07b95` +covers 68 engine-step records, 19 compiled-engine references, and three +per-op cases. It follows removal of eager vLLM 0.24.0 FP8-block timings +and explicit reuse of the graph-timed 0.25.0 GEMM table. Declared reuse +also fills missing GEMM shapes for the other retained precisions. + +The three per-op cases change only 12 QKV/projection GEMM latency values; +other per-op latencies, energies, and source labels are unchanged. For +MiniMax-M2.5 (B200, ISL 1024, OSL 2), context QKV GEMM changes from +11.926155 to 0.948021 ms and generation QKV GEMM from 16.582272 to +0.565109 ms. Static, mixed-step, aggregated/disaggregated, chunked-prefill, +and imbalance-scale references inherit these data changes. + +Only records implicated by the failed golden comparisons were refreshed, +using `pin_goldens.py --refresh`. Each refreshed record retains its source +commit in `post_freeze_pins`; test matrices and tolerances are unchanged. +These are prediction-regression baselines, not whole-model silicon validation. + ## Engine-Step Benchmark Historical Python-vs-Rust speedup numbers (dated + commit-stamped) live in diff --git a/crates/core/parity_tests/perfmodel/goldens/compile_engine.json b/crates/core/parity_tests/perfmodel/goldens/compile_engine.json index 0793a72c6f..fa3a1e8d71 100644 --- a/crates/core/parity_tests/perfmodel/goldens/compile_engine.json +++ b/crates/core/parity_tests/perfmodel/goldens/compile_engine.json @@ -16,15 +16,15 @@ "post_freeze_pins": { "chunked_prefill::ctx300_gen7_isl1000_osl64_prefix100::mixed_step": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "chunked_prefill::ctx512_gen4_isl4096_osl128_prefix0::mixed_step": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "chunked_prefill::ctx512_gen4_isl4096_osl128_prefix256::mixed_step": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "gpt-oss-20b-b200-trtllm-isl1024-osl2::decode_step": { "engine": "rust", @@ -44,19 +44,19 @@ }, "imbalance_scale::decode_step": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "imbalance_scale::mixed_step": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "imbalance_scale::static_ctx": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "imbalance_scale::static_gen": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "kimi-k25-b200-sglang-isl1024-osl2::decode_step": { "engine": "rust", @@ -108,51 +108,51 @@ }, "minimax-m25-b200-vllm-isl1024-osl2::decode_step": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-vllm-isl1024-osl2::mixed_step": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-vllm-isl1024-osl2::static_ctx": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-vllm-isl1024-osl2::static_gen": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-vllm-sampled-prefix::decode_step": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-vllm-sampled-prefix::mixed_step": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-vllm-sampled-prefix::static_ctx": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-vllm-sampled-prefix::static_gen": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m27-b200-vllm-isl1024-osl2::decode_step": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m27-b200-vllm-isl1024-osl2::mixed_step": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m27-b200-vllm-isl1024-osl2::static_ctx": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m27-b200-vllm-isl1024-osl2::static_gen": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "nemotron-nas-b200-trtllm-isl1024-osl2::decode_step": { "engine": "rust", @@ -228,17 +228,17 @@ } }, "references": { - "chunked_prefill::ctx300_gen7_isl1000_osl64_prefix100::mixed_step": 43.58332153821506, - "chunked_prefill::ctx512_gen4_isl4096_osl128_prefix0::mixed_step": 46.03136681155293, - "chunked_prefill::ctx512_gen4_isl4096_osl128_prefix256::mixed_step": 46.029415749083185, + "chunked_prefill::ctx300_gen7_isl1000_osl64_prefix100::mixed_step": 16.081171352808852, + "chunked_prefill::ctx512_gen4_isl4096_osl128_prefix0::mixed_step": 17.664446149246828, + "chunked_prefill::ctx512_gen4_isl4096_osl128_prefix256::mixed_step": 17.662495086777085, "gpt-oss-20b-b200-trtllm-isl1024-osl2::decode_step": 2.213361158220323, "gpt-oss-20b-b200-trtllm-isl1024-osl2::mixed_step": 11.539696083103436, "gpt-oss-20b-b200-trtllm-isl1024-osl2::static_ctx": 11.215364152344467, "gpt-oss-20b-b200-trtllm-isl1024-osl2::static_gen": 2.213362545722157, - "imbalance_scale::decode_step": 39.01752053715316, - "imbalance_scale::mixed_step": 53.71932103484562, - "imbalance_scale::static_ctx": 43.88750156247328, - "imbalance_scale::static_gen": 39.01750961288222, + "imbalance_scale::decode_step": 6.609541889454197, + "imbalance_scale::mixed_step": 26.06666001428854, + "imbalance_scale::static_ctx": 23.110060488022473, + "imbalance_scale::static_gen": 6.60953096518326, "kimi-k25-b200-sglang-isl1024-osl2::decode_step": 6.491702647659921, "kimi-k25-b200-sglang-isl1024-osl2::mixed_step": 40.250282696160745, "kimi-k25-b200-sglang-isl1024-osl2::static_ctx": 39.65105076373316, @@ -251,18 +251,18 @@ "minimax-m25-b200-sglang-isl1024-osl2::mixed_step": 28.09828061363558, "minimax-m25-b200-sglang-isl1024-osl2::static_ctx": 25.330444037457216, "minimax-m25-b200-sglang-isl1024-osl2::static_gen": 7.412970250723051, - "minimax-m25-b200-vllm-isl1024-osl2::decode_step": 39.36457150698851, - "minimax-m25-b200-vllm-isl1024-osl2::mixed_step": 52.85520178157888, - "minimax-m25-b200-vllm-isl1024-osl2::static_ctx": 42.67633133937119, - "minimax-m25-b200-vllm-isl1024-osl2::static_gen": 39.36455865490505, - "minimax-m25-b200-vllm-sampled-prefix::decode_step": 35.194031677138476, - "minimax-m25-b200-vllm-sampled-prefix::mixed_step": 60.67073561230981, - "minimax-m25-b200-vllm-sampled-prefix::static_ctx": 68.94160745246145, - "minimax-m25-b200-vllm-sampled-prefix::static_gen": 140.77605550543933, - "minimax-m27-b200-vllm-isl1024-osl2::decode_step": 39.36457150698851, - "minimax-m27-b200-vllm-isl1024-osl2::mixed_step": 52.85520178157888, - "minimax-m27-b200-vllm-isl1024-osl2::static_ctx": 42.67633133937119, - "minimax-m27-b200-vllm-isl1024-osl2::static_gen": 39.36455865490505, + "minimax-m25-b200-vllm-isl1024-osl2::decode_step": 6.956592859289543, + "minimax-m25-b200-vllm-isl1024-osl2::mixed_step": 25.202540761021798, + "minimax-m25-b200-vllm-isl1024-osl2::static_ctx": 21.898890264920386, + "minimax-m25-b200-vllm-isl1024-osl2::static_gen": 6.956580007206089, + "minimax-m25-b200-vllm-sampled-prefix::decode_step": 7.81648403057622, + "minimax-m25-b200-vllm-sampled-prefix::mixed_step": 33.61569835698758, + "minimax-m25-b200-vllm-sampled-prefix::static_ctx": 48.247495936383125, + "minimax-m25-b200-vllm-sampled-prefix::static_gen": 31.26586491919031, + "minimax-m27-b200-vllm-isl1024-osl2::decode_step": 6.956592859289543, + "minimax-m27-b200-vllm-isl1024-osl2::mixed_step": 25.202540761021798, + "minimax-m27-b200-vllm-isl1024-osl2::static_ctx": 21.898890264920386, + "minimax-m27-b200-vllm-isl1024-osl2::static_gen": 6.956580007206089, "nemotron-nas-b200-trtllm-isl1024-osl2::decode_step": 4.86243033208606, "nemotron-nas-b200-trtllm-isl1024-osl2::mixed_step": 24.98626430220314, "nemotron-nas-b200-trtllm-isl1024-osl2::static_ctx": 24.91795031094951, diff --git a/crates/core/parity_tests/perfmodel/goldens/engine_step.json b/crates/core/parity_tests/perfmodel/goldens/engine_step.json index 8398ef540d..848ff04bee 100644 --- a/crates/core/parity_tests/perfmodel/goldens/engine_step.json +++ b/crates/core/parity_tests/perfmodel/goldens/engine_step.json @@ -3,84 +3,84 @@ "deepseek-r1-b200-vllm-isl1024-osl2": { "agg": { "values": { - "agg_request": 226.97449032341171, - "agg_tpot": 80.97623074042903, - "agg_ttft": 145.99825958298268 + "agg_request": 171.23574164975162, + "agg_tpot": 54.746231364589, + "agg_ttft": 116.48951028516262 } }, "disagg": { "values": { - "disagg_request": 174.955, - "disagg_tpot": 27.978, - "disagg_ttft": 146.977 + "disagg_request": 118.876, + "disagg_tpot": 11.996, + "disagg_ttft": 106.88 } }, "mixed": { "values": { - "mixed_step": 80.97623074042902 + "mixed_step": 54.74623136458901 } }, "static": { "values": { - "static_ctx": 74.231, - "static_gen": 37.318, - "static_total": 111.54899999999999 + "static_ctx": 53.98, + "static_gen": 10.172, + "static_total": 64.152 } } }, "deepseek-r1-gb200-vllm-scan-coverage": { "agg": { "values": { - "agg_request": 295.22849734948784, - "agg_tpot": 113.09576345858251, - "agg_ttft": 182.1327338909053 + "agg_request": 190.65534207704624, + "agg_tpot": 63.88486685978646, + "agg_ttft": 126.77047521725977 } }, "disagg": { "values": { - "disagg_request": 238.204, - "disagg_tpot": 52.883, - "disagg_ttft": 185.321 + "disagg_request": 137.51, + "disagg_tpot": 12.862, + "disagg_ttft": 124.648 } }, "mixed": { "values": { - "mixed_step": 113.0957634585825 + "mixed_step": 63.88486685978646 } }, "static": { "values": { - "static_ctx": 93.596, - "static_gen": 69.934, - "static_total": 163.53 + "static_ctx": 62.954, + "static_gen": 10.556, + "static_total": 73.51 } } }, "deepseek-v3-b200-vllm-shape-prefix-heavy": { "agg": { "values": { - "agg_request": 751.4422001436218, - "agg_tpot": 40.28540345952531, - "agg_ttft": 147.16114825074214 + "agg_request": 358.3741439734639, + "agg_tpot": 16.048116334702787, + "agg_ttft": 117.65239895292213 } }, "disagg": { "values": { - "disagg_request": 569.68, - "disagg_tpot": 28.044, - "disagg_ttft": 149.02 + "disagg_request": 289.855, + "disagg_tpot": 12.062, + "disagg_ttft": 108.925 } }, "mixed": { "values": { - "mixed_step": 82.00990955621523 + "mixed_step": 55.779910180375225 } }, "static": { "values": { - "static_ctx": 75.263, - "static_gen": 561.048, - "static_total": 636.311 + "static_ctx": 55.013, + "static_gen": 153.867, + "static_total": 208.88 } } }, @@ -129,84 +129,84 @@ "deepseek-v32-b200-vllm-empirical": { "agg": { "values": { - "agg_request": 437.9756148647986, - "agg_tpot": 180.27087758343464, - "agg_ttft": 257.704737281364 + "agg_request": 384.7981398952146, + "agg_tpot": 155.24618348010097, + "agg_ttft": 229.5519564151136 } }, "disagg": { "values": { - "disagg_request": 359.646, - "disagg_tpot": 35.108, - "disagg_ttft": 324.538 + "disagg_request": 307.169, + "disagg_tpot": 19.104, + "disagg_ttft": 288.065 } }, "mixed": { "values": { - "mixed_step": 180.27087758343467 + "mixed_step": 155.24618348010097 } }, "static": { "values": { - "static_ctx": 163.908, - "static_gen": 42.809, - "static_total": 206.71699999999998 + "static_ctx": 145.487, + "static_gen": 15.165, + "static_total": 160.652 } } }, "deepseek-v32-b200-vllm-isl1024-osl2": { "agg": { "values": { - "agg_request": 437.97918696457896, - "agg_tpot": 180.27255857156658, - "agg_ttft": 257.7066283930124 + "agg_request": 385.0603863780549, + "agg_tpot": 155.3695935896729, + "agg_ttft": 229.690792788382 } }, "disagg": { "values": { - "disagg_request": 359.586, - "disagg_tpot": 35.107, - "disagg_ttft": 324.479 + "disagg_request": 306.977, + "disagg_tpot": 19.132, + "disagg_ttft": 287.845 } }, "mixed": { "values": { - "mixed_step": 180.27255857156655 + "mixed_step": 155.3695935896729 } }, "static": { "values": { - "static_ctx": 163.878, - "static_gen": 42.809, - "static_total": 206.68699999999998 + "static_ctx": 145.377, + "static_gen": 15.191, + "static_total": 160.568 } } }, "deepseek-v4-flash-b200-vllm-isl1024-osl2": { "agg": { "values": { - "agg_request": 134.62931630314552, - "agg_tpot": 45.14320767206848, - "agg_ttft": 89.48610863107703 + "agg_request": 93.47255476257988, + "agg_tpot": 25.775319888272882, + "agg_ttft": 67.69723487430699 } }, "disagg": { "values": { - "disagg_request": 93.983, - "disagg_tpot": 19.963, - "disagg_ttft": 74.02 + "disagg_request": 50.577, + "disagg_tpot": 5.887, + "disagg_ttft": 44.69 } }, "mixed": { "values": { - "mixed_step": 45.143207672068485 + "mixed_step": 25.77531988827288 } }, "static": { "values": { - "static_ctx": 37.383, - "static_gen": 26.53, - "static_total": 63.913000000000004 + "static_ctx": 22.571, + "static_gen": 5.241, + "static_total": 27.812 } } }, @@ -571,9 +571,9 @@ }, "disagg": { "values": { - "disagg_request": 49.038, - "disagg_tpot": 7.084, - "disagg_ttft": 41.954 + "disagg_request": 49.402, + "disagg_tpot": 7.233, + "disagg_ttft": 42.169 } }, "mixed": { @@ -583,37 +583,37 @@ }, "static": { "values": { - "static_ctx": 21.189, - "static_gen": 4.522, - "static_total": 25.711 + "static_ctx": 21.297, + "static_gen": 4.619, + "static_total": 25.916 } } }, "mimo-v2-flash-b200-vllm-hybrid-attn-xshape": { "agg": { "values": { - "agg_request": 139.04908593658314, - "agg_tpot": 45.10545220545088, - "agg_ttft": 93.94363373113225 + "agg_request": 93.51656956202063, + "agg_tpot": 23.678385676245004, + "agg_ttft": 69.83818388577563 } }, "disagg": { "values": { - "disagg_request": 99.905, - "disagg_tpot": 23.45, - "disagg_ttft": 76.455 + "disagg_request": 50.373, + "disagg_tpot": 6.149, + "disagg_ttft": 44.224 } }, "mixed": { "values": { - "mixed_step": 45.10545220545086 + "mixed_step": 23.678385676245007 } }, "static": { "values": { - "static_ctx": 38.614, - "static_gen": 29.587, - "static_total": 68.201 + "static_ctx": 22.335, + "static_gen": 4.416, + "static_total": 26.751 } } }, @@ -648,112 +648,112 @@ "minimax-m25-b200-vllm-empirical": { "agg": { "values": { - "agg_request": 168.1374758861998, - "agg_tpot": 52.86469453468226, - "agg_ttft": 115.27278135151754 + "agg_request": 107.93990632676808, + "agg_tpot": 24.53642650671439, + "agg_ttft": 83.40347982005369 } }, "disagg": { "values": { - "disagg_request": 114.909, - "disagg_tpot": 30.379, - "disagg_ttft": 84.53 + "disagg_request": 51.654, + "disagg_tpot": 7.696, + "disagg_ttft": 43.958 } }, "mixed": { "values": { - "mixed_step": 52.86469453468228 + "mixed_step": 24.536426506714385 } }, "static": { "values": { - "static_ctx": 42.692, - "static_gen": 39.318, - "static_total": 82.00999999999999 + "static_ctx": 22.201, + "static_gen": 8.172, + "static_total": 30.373 } } }, "minimax-m25-b200-vllm-isl1024-osl2": { "agg": { "values": { - "agg_request": 168.1173037858551, - "agg_tpot": 52.85520178157888, - "agg_ttft": 115.26210200427624 + "agg_request": 109.35539911717132, + "agg_tpot": 25.202540761021798, + "agg_ttft": 84.15285835614952 } }, "disagg": { "values": { - "disagg_request": 114.936, - "disagg_tpot": 30.437, - "disagg_ttft": 84.499 + "disagg_request": 51.619, + "disagg_tpot": 8.259, + "disagg_ttft": 43.36 } }, "mixed": { "values": { - "mixed_step": 52.85520178157888 + "mixed_step": 25.202540761021798 } }, "static": { "values": { - "static_ctx": 42.676, - "static_gen": 39.365, - "static_total": 82.041 + "static_ctx": 21.899, + "static_gen": 6.957, + "static_total": 28.856 } } }, "minimax-m25-b200-vllm-sampled-prefix": { "agg": { "values": { - "agg_request": 305.7590467617681, - "agg_tpot": 45.409023885055305, - "agg_ttft": 124.12295122154688 + "agg_request": 166.24926626789278, + "agg_tpot": 18.148929624879287, + "agg_ttft": 93.65354776837562 } }, "disagg": { "values": { - "disagg_request": 221.813, - "disagg_tpot": 30.562, - "disagg_ttft": 99.565 + "disagg_request": 93.863, + "disagg_tpot": 8.383, + "disagg_ttft": 60.331 } }, "mixed": { "values": { - "mixed_step": 60.67073561230981 + "mixed_step": 33.61569835698758 } }, "static": { "values": { - "static_ctx": 68.942, - "static_gen": 140.776, - "static_total": 209.71800000000002 + "static_ctx": 48.247, + "static_gen": 31.266, + "static_total": 79.513 } } }, "minimax-m25-b200-vllm-shape-decode-heavy": { "agg": { "values": { - "agg_request": 2335.0429355970573, - "agg_tpot": 35.39861468775233, - "agg_ttft": 104.93021026866046 + "agg_request": 581.4586824211743, + "agg_tpot": 8.041279057956958, + "agg_ttft": 74.85810176988599 } }, "disagg": { "values": { - "disagg_request": 1989.857, - "disagg_tpot": 30.443, - "disagg_ttft": 71.948 + "disagg_request": 548.23, + "disagg_tpot": 8.264, + "disagg_ttft": 27.598 } }, "mixed": { "values": { - "mixed_step": 43.67129801658708 + "mixed_step": 16.94053490656533 } }, "static": { "values": { - "static_ctx": 36.337, - "static_gen": 2471.865, - "static_total": 2508.2019999999998 + "static_ctx": 13.938, + "static_gen": 430.162, + "static_total": 444.09999999999997 } } }, @@ -800,28 +800,28 @@ "minimax-m27-b200-vllm-isl1024-osl2": { "agg": { "values": { - "agg_request": 168.1173037858551, - "agg_tpot": 52.85520178157888, - "agg_ttft": 115.26210200427624 + "agg_request": 109.35539911717132, + "agg_tpot": 25.202540761021798, + "agg_ttft": 84.15285835614952 } }, "disagg": { "values": { - "disagg_request": 114.936, - "disagg_tpot": 30.437, - "disagg_ttft": 84.499 + "disagg_request": 51.619, + "disagg_tpot": 8.259, + "disagg_ttft": 43.36 } }, "mixed": { "values": { - "mixed_step": 52.85520178157888 + "mixed_step": 25.202540761021798 } }, "static": { "values": { - "static_ctx": 42.676, - "static_gen": 39.365, - "static_total": 82.041 + "static_ctx": 21.899, + "static_gen": 6.957, + "static_total": 28.856 } } }, @@ -856,84 +856,84 @@ "minimax-m3-b200-vllm-hybrid-policy-balanced": { "agg": { "values": { - "agg_request": 141.97580829173353, - "agg_tpot": 41.400380372580486, - "agg_ttft": 100.57542791915304 + "agg_request": 141.9350722523982, + "agg_tpot": 41.381210471716784, + "agg_ttft": 100.55386178068139 } }, "disagg": { "values": { - "disagg_request": 89.972, + "disagg_request": 89.963, "disagg_tpot": 11.46, - "disagg_ttft": 78.512 + "disagg_ttft": 78.503 } }, "mixed": { "values": { - "mixed_step": 41.400380372580486 + "mixed_step": 41.381210471716784 } }, "static": { "values": { - "static_ctx": 39.653, + "static_ctx": 39.649, "static_gen": 7.943, - "static_total": 47.596 + "static_total": 47.592 } } }, "minimax-m3-b200-vllm-hybrid-policy-off": { "agg": { "values": { - "agg_request": 141.97580829173353, - "agg_tpot": 41.400380372580486, - "agg_ttft": 100.57542791915304 + "agg_request": 141.9350722523982, + "agg_tpot": 41.381210471716784, + "agg_ttft": 100.55386178068139 } }, "disagg": { "values": { - "disagg_request": 89.972, + "disagg_request": 89.963, "disagg_tpot": 11.46, - "disagg_ttft": 78.512 + "disagg_ttft": 78.503 } }, "mixed": { "values": { - "mixed_step": 41.400380372580486 + "mixed_step": 41.381210471716784 } }, "static": { "values": { - "static_ctx": 39.653, + "static_ctx": 39.649, "static_gen": 7.943, - "static_total": 47.596 + "static_total": 47.592 } } }, "minimax-m3-b200-vllm-hybrid-xop": { "agg": { "values": { - "agg_request": 141.97580829173353, - "agg_tpot": 41.400380372580486, - "agg_ttft": 100.57542791915304 + "agg_request": 141.9350722523982, + "agg_tpot": 41.381210471716784, + "agg_ttft": 100.55386178068139 } }, "disagg": { "values": { - "disagg_request": 89.972, + "disagg_request": 89.963, "disagg_tpot": 11.46, - "disagg_ttft": 78.512 + "disagg_ttft": 78.503 } }, "mixed": { "values": { - "mixed_step": 41.400380372580486 + "mixed_step": 41.381210471716784 } }, "static": { "values": { - "static_ctx": 39.653, + "static_ctx": 39.649, "static_gen": 7.943, - "static_total": 47.596 + "static_total": 47.592 } } }, @@ -1575,28 +1575,28 @@ "qwen35-27b-b200-vllm-isl1024-osl2": { "agg": { "values": { - "agg_request": 94.36257936271598, - "agg_tpot": 17.30003734716046, - "agg_ttft": 77.06254201555552 + "agg_request": 95.2575458007199, + "agg_tpot": 17.721198023868183, + "agg_ttft": 77.53634777685171 } }, "disagg": { "values": { - "disagg_request": 37.261, - "disagg_tpot": 4.906, - "disagg_ttft": 32.355 + "disagg_request": 39.196, + "disagg_tpot": 5.669, + "disagg_ttft": 33.527 } }, "mixed": { "values": { - "mixed_step": 17.300037347160462 + "mixed_step": 17.721198023868187 } }, "static": { "values": { - "static_ctx": 16.34, - "static_gen": 4.491, - "static_total": 20.831 + "static_ctx": 16.932, + "static_gen": 4.617, + "static_total": 21.549 } } }, @@ -1666,9 +1666,9 @@ }, "disagg": { "values": { - "disagg_request": 61.748, - "disagg_tpot": 6.457, - "disagg_ttft": 55.291 + "disagg_request": 62.364, + "disagg_tpot": 6.523, + "disagg_ttft": 55.841 } }, "mixed": { @@ -1678,9 +1678,9 @@ }, "static": { "values": { - "static_ctx": 27.925, - "static_gen": 4.911, - "static_total": 32.836 + "static_ctx": 28.203, + "static_gen": 4.979, + "static_total": 33.182 } } }, @@ -1843,51 +1843,51 @@ "post_freeze_pins": { "deepseek-r1-b200-vllm-isl1024-osl2::agg": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "deepseek-r1-b200-vllm-isl1024-osl2::disagg": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "deepseek-r1-b200-vllm-isl1024-osl2::mixed": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "deepseek-r1-b200-vllm-isl1024-osl2::static": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "deepseek-r1-gb200-vllm-scan-coverage::agg": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "deepseek-r1-gb200-vllm-scan-coverage::disagg": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "deepseek-r1-gb200-vllm-scan-coverage::mixed": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "deepseek-r1-gb200-vllm-scan-coverage::static": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "deepseek-v3-b200-vllm-shape-prefix-heavy::agg": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "deepseek-v3-b200-vllm-shape-prefix-heavy::disagg": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "deepseek-v3-b200-vllm-shape-prefix-heavy::mixed": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "deepseek-v3-b200-vllm-shape-prefix-heavy::static": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "deepseek-v3-b200-vllm-sol::mixed": { "engine": "rust", @@ -1915,51 +1915,51 @@ }, "deepseek-v32-b200-vllm-empirical::agg": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "deepseek-v32-b200-vllm-empirical::disagg": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "deepseek-v32-b200-vllm-empirical::mixed": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "deepseek-v32-b200-vllm-empirical::static": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "deepseek-v32-b200-vllm-isl1024-osl2::agg": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "deepseek-v32-b200-vllm-isl1024-osl2::disagg": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "deepseek-v32-b200-vllm-isl1024-osl2::mixed": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "deepseek-v32-b200-vllm-isl1024-osl2::static": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "deepseek-v4-flash-b200-vllm-isl1024-osl2::agg": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "deepseek-v4-flash-b200-vllm-isl1024-osl2::disagg": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "deepseek-v4-flash-b200-vllm-isl1024-osl2::mixed": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "deepseek-v4-flash-b200-vllm-isl1024-osl2::static": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "dsv4-flash-b200-sglang-cp8-reuse::cp_static_ctx": { "engine": "rust", @@ -2187,7 +2187,7 @@ }, "llama4-scout-b200-vllm-isl1024-osl2::disagg": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "llama4-scout-b200-vllm-isl1024-osl2::mixed": { "engine": "rust", @@ -2195,23 +2195,23 @@ }, "llama4-scout-b200-vllm-isl1024-osl2::static": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "mimo-v2-flash-b200-vllm-hybrid-attn-xshape::agg": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "mimo-v2-flash-b200-vllm-hybrid-attn-xshape::disagg": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "mimo-v2-flash-b200-vllm-hybrid-attn-xshape::mixed": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "mimo-v2-flash-b200-vllm-hybrid-attn-xshape::static": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-sglang-isl1024-osl2::agg": { "engine": "rust", @@ -2231,67 +2231,67 @@ }, "minimax-m25-b200-vllm-empirical::agg": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-vllm-empirical::disagg": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-vllm-empirical::mixed": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-vllm-empirical::static": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-vllm-isl1024-osl2::agg": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-vllm-isl1024-osl2::disagg": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-vllm-isl1024-osl2::mixed": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-vllm-isl1024-osl2::static": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-vllm-sampled-prefix::agg": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-vllm-sampled-prefix::disagg": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-vllm-sampled-prefix::mixed": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-vllm-sampled-prefix::static": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-vllm-shape-decode-heavy::agg": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-vllm-shape-decode-heavy::disagg": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-vllm-shape-decode-heavy::mixed": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-vllm-shape-decode-heavy::static": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-nvfp4-h200-vllm-hybrid-balanced-miss::agg": { "engine": "rust", @@ -2327,19 +2327,19 @@ }, "minimax-m27-b200-vllm-isl1024-osl2::agg": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m27-b200-vllm-isl1024-osl2::disagg": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m27-b200-vllm-isl1024-osl2::mixed": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m27-b200-vllm-isl1024-osl2::static": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m3-b200-sglang-hybrid-xop::agg": { "engine": "rust", @@ -2359,51 +2359,51 @@ }, "minimax-m3-b200-vllm-hybrid-policy-balanced::agg": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m3-b200-vllm-hybrid-policy-balanced::disagg": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m3-b200-vllm-hybrid-policy-balanced::mixed": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m3-b200-vllm-hybrid-policy-balanced::static": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m3-b200-vllm-hybrid-policy-off::agg": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m3-b200-vllm-hybrid-policy-off::disagg": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m3-b200-vllm-hybrid-policy-off::mixed": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m3-b200-vllm-hybrid-policy-off::static": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m3-b200-vllm-hybrid-xop::agg": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m3-b200-vllm-hybrid-xop::disagg": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m3-b200-vllm-hybrid-xop::mixed": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m3-b200-vllm-hybrid-xop::static": { "engine": "rust", - "git_head": "760b6c92edc00389ff79938b62b028c01bcfefdd" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "nemotron-h-56b-b200-sglang-isl1024-osl2::agg": { "engine": "rust", @@ -2767,19 +2767,19 @@ }, "qwen35-27b-b200-vllm-isl1024-osl2::agg": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "qwen35-27b-b200-vllm-isl1024-osl2::disagg": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "qwen35-27b-b200-vllm-isl1024-osl2::mixed": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "qwen35-27b-b200-vllm-isl1024-osl2::static": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "qwen35-397b-a17b-b200-sglang-isl1024-osl2::agg": { "engine": "rust", @@ -2819,7 +2819,7 @@ }, "qwen35-397b-a17b-b200-vllm-isl1024-osl2::disagg": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "qwen35-397b-a17b-b200-vllm-isl1024-osl2::mixed": { "engine": "rust", @@ -2827,7 +2827,7 @@ }, "qwen35-397b-a17b-b200-vllm-isl1024-osl2::static": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "qwen35-397b-a17b-gb200-sglang-0514-tp16-scan-coverage::agg": { "engine": "rust", diff --git a/crates/core/parity_tests/perfmodel/goldens/per_op.json b/crates/core/parity_tests/perfmodel/goldens/per_op.json index 17b57951fd..45f59769c2 100644 --- a/crates/core/parity_tests/perfmodel/goldens/per_op.json +++ b/crates/core/parity_tests/perfmodel/goldens/per_op.json @@ -533,12 +533,12 @@ }, "context_proj_gemm": { "energy_wms": 0.0, - "latency_ms": 10.640523235003155, + "latency_ms": 0.8412160153190296, "source": "silicon" }, "context_qkv_gemm": { "energy_wms": 0.0, - "latency_ms": 11.92615516980489, + "latency_ms": 0.9480213150382042, "source": "silicon" }, "context_router_gemm": { @@ -600,12 +600,12 @@ }, "generation_proj_gemm": { "energy_wms": 0.0, - "latency_ms": 16.94336028893789, + "latency_ms": 0.5525440114239852, "source": "silicon" }, "generation_qkv_gemm": { "energy_wms": 0.0, - "latency_ms": 16.582271695137024, + "latency_ms": 0.5651093249519665, "source": "silicon" }, "generation_router_gemm": { @@ -664,12 +664,12 @@ }, "context_proj_gemm": { "energy_wms": 0.0, - "latency_ms": 12.246901353200276, + "latency_ms": 1.4360852589209876, "source": "silicon" }, "context_qkv_gemm": { "energy_wms": 0.0, - "latency_ms": 11.719487468401592, + "latency_ms": 1.8361920466025672, "source": "silicon" }, "context_router_gemm": { @@ -731,12 +731,12 @@ }, "generation_proj_gemm": { "energy_wms": 0.0, - "latency_ms": 59.742209593455, + "latency_ms": 3.770922740300496, "source": "silicon" }, "generation_qkv_gemm": { "energy_wms": 0.0, - "latency_ms": 57.26617844899495, + "latency_ms": 3.727274715900421, "source": "silicon" }, "generation_router_gemm": { @@ -795,12 +795,12 @@ }, "context_proj_gemm": { "energy_wms": 0.0, - "latency_ms": 10.640523235003155, + "latency_ms": 0.8412160153190296, "source": "silicon" }, "context_qkv_gemm": { "energy_wms": 0.0, - "latency_ms": 11.92615516980489, + "latency_ms": 0.9480213150382042, "source": "silicon" }, "context_router_gemm": { @@ -862,12 +862,12 @@ }, "generation_proj_gemm": { "energy_wms": 0.0, - "latency_ms": 16.94336028893789, + "latency_ms": 0.5525440114239852, "source": "silicon" }, "generation_qkv_gemm": { "energy_wms": 0.0, - "latency_ms": 16.582271695137024, + "latency_ms": 0.5651093249519665, "source": "silicon" }, "generation_router_gemm": { @@ -1178,15 +1178,15 @@ }, "minimax-m25-b200-vllm-isl1024-osl2": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m25-b200-vllm-sampled-prefix": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "minimax-m27-b200-vllm-isl1024-osl2": { "engine": "rust", - "git_head": "eb4f61cb5d11f17898d7a0f1fd47def268445a6b" + "git_head": "0a51476b6ab90bc0e475bd41d4b1c7abbef07b95" }, "nemotron-nas-b200-trtllm-isl1024-osl2": { "engine": "rust", diff --git a/docs/ci.md b/docs/ci.md index b5d6bc47cc..e9a9a06895 100644 --- a/docs/ci.md +++ b/docs/ci.md @@ -239,6 +239,16 @@ are 2% relative and 0.0001 ms absolute. Missing, duplicate, failed, nonfinite, nonpositive, or out-of-tolerance results fail. Intentional modeling changes need explained before/after evidence; do not refresh goldens merely to pass CI. +The FP8-block data correction in PR #244 changes only the MiniMax cases to +enable declared reuse: their vLLM 0.24.0 primary data no longer contains +invalid eager FP8-block measurements, so corrected GEMMs come from 0.25.0. +With the original data, enabling reuse reproduces all four old baselines +exactly. With corrected data, the prefill baselines change from +42.676331 / 116.409355 ms to 21.898890 / 98.968257 ms, and decode from +39.364559 / 7396.437641 ms to 6.956580 / 2192.966039 ms (short/long cases). +The four Qwen baselines and all tolerances remain unchanged. These are +prediction-stability values, not measured whole-model accuracy. + The broader [prediction comparison](../python/aisimulate/tools/prediction_regression_gate/report.py) reports numerical drift, gains, and added/removed rows for review. It blocks previously working cases becoming broken. If the comparison base predates the diff --git a/docs/data/pr244/README.md b/docs/data/pr244/README.md new file mode 100644 index 0000000000..66a8a31f74 --- /dev/null +++ b/docs/data/pr244/README.md @@ -0,0 +1,106 @@ +# PR #244 collection evidence + +This non-packaged bundle makes the vLLM 0.25.0 collection claims reproducible +without a GPU, host paths, usernames, scheduler IDs, or raw diagnostic logs. +It is outside the Python package's source/data tree. + +## Verify + +From the repository root, after `uv sync --project python/aisimulate --extra dev`: + +```sh +python/aisimulate/.venv/bin/python scripts/audit_pr244_data.py +``` + +The check recomputes all case counts and summaries of the 85 shipped parquet +files. It checks each file's hash and row count against the original collection +report, then compares the result with `manifest.json`. No network or historical +Git objects are needed for this check. Use `--write` only when intentionally +regenerating the manifest after reviewing a change. +Both check and write modes reject tables with null cells, duplicate physical +keys, or nonfinite/nonpositive latency, even when the original report's hash +and row count match. Rejection leaves the existing manifest unchanged. + +| System | Tables | Rows | Retained cases | Done | Failed | Unattempted | +| --- | ---: | ---: | ---: | ---: | ---: | ---: | +| B300 | 17 | 420,111 | 350,380 | 339,641 | 10,739 | 0 | +| GB200 | 17 | 420,134 | 350,380 | 339,614 | 10,766 | 0 | +| GB300 | 17 | 420,226 | 350,380 | 339,681 | 10,699 | 0 | +| H100 | 17 | 353,449 | 294,319 | 290,824 | 3,495 | 0 | +| H200 | 17 | 353,658 | 294,391 | 290,997 | 3,394 | 0 | + +All 85 tables have zero reported null cells, duplicate physical keys, and +nonfinite or nonpositive latency values. + +Cases and rows are different units: a single MoE/GDN case can emit many rows. +Zero unattempted cases means every **retained** case reached a final checkpoint +outcome. It does not mean every possible shape succeeded or that capability +filtering retained every raw candidate. Failed cases remain failed; they are +not converted into timing rows or silently counted as successful coverage. + +## Files and provenance + +- `manifest.json`: per-system and per-attempt counts; per-table SHA-256, rows, + physical key columns, Arrow types, numeric shape ranges, categorical domains, + distinct counts, latency ranges in **milliseconds**, and anomaly counts + (nulls, duplicate physical keys, nonfinite/nonpositive latency). Sidecar hashes + bind the summary to the packaged provenance. Physical row identity includes + `kernel_source`; engine lookup keys can be narrower. +- `case-outcomes.json.gz`: deterministic gzip of JSON. `plans[sha256]` contains + the complete sorted retained task-ID list. The digest is SHA-256 of the IDs + joined by newline, with **no final newline**. Each system's attempt references + one plan and an equally long `outcomes` string: `D` = done, `F` = failed, + `U` = unattempted. Zip the IDs with that string to inspect any failed shape. + Plans shared across attempts/systems are stored once. Original report/archive + hashes and reported table hashes/counts are retained. +- Source: the final collection reports and failure archives at repository commit + [`1f29ee459882796db312b8a288c040b83edbb211`](https://github.com/ai-dynamo/aisimulate/commit/1f29ee459882796db312b8a288c040b83edbb211), + before diagnostic cleanup. Only final top-level checkpoints are extracted; + earlier attempts and retry logs are not summed into the results. +- B300's archived standalone case-plan object was absent. Its exact retained IDs + were recovered from the final checkpoint union, then checked against **both** + the independently recorded plan digest and expected count for every attempt. + This is recorded as `verified_checkpoint_union`; other systems use their + `archived_case_plan`. A mismatch stops extraction instead of inferring zero + unattempted cases from the observed outcomes alone. +- Collector revision: `cbaf51b64fa460e5ec6146bde407a4c64958212d`. + Actual vLLM runtime revision: `dd10e03f95f94edbea1975c67ace3a35ec9a8a40`. + The runtime revision differs from the v0.25.0 tag's commit. + +To reproduce the one-time sanitized extraction when the source Git objects are +available, run the same command with `--extract-cases --write`. Extraction +allowlists only case IDs, outcomes, counts, and hashes; it does not copy arbitrary +report fields, paths, error messages, or runtime environment dictionaries. +The evidence supports collection coverage and table integrity, not diagnosis of +each failure or end-to-end prediction accuracy. + +## Why table-wide GEMM reuse is intentional + +The six 0.24.0 declarations (B200/B300/GB200/GB300/H100/H200) fill **every** +unshadowed 0.25.0 GEMM key. They add 36,408 FP8-block keys, 1,776 BF16 keys, +and 1,776 FP8 keys per system, plus 1,776 NVFP4 keys on Blackwell. Retained +0.24.0 primary rows always win, including their exact measured latency. + +The non-FP8-block donor rows record these selected kernels. An immutable source +comparison between vLLM +[`v0.24.0 / ee0da84ab9e04ac7610e28580af62c365e898389`](https://github.com/vllm-project/vllm/tree/ee0da84ab9e04ac7610e28580af62c365e898389) +and the +[actual 0.25.0 runtime](https://github.com/vllm-project/vllm/tree/dd10e03f95f94edbea1975c67ace3a35ec9a8a40) +found: + +| Precision | Recorded kernel | Source comparison | +| --- | --- | --- | +| BF16 | `torch.nn.functional.linear` | `RowParallelLinear` and `UnquantizedLinearMethod` in [`layers/linear.py`](https://github.com/vllm-project/vllm/blob/dd10e03f95f94edbea1975c67ace3a35ec9a8a40/vllm/model_executor/layers/linear.py) have identical Python ASTs. | +| FP8 | `CutlassFP8ScaledMMLinearKernel` | [`kernels/linear/scaled_mm/cutlass.py`](https://github.com/vllm-project/vllm/blob/dd10e03f95f94edbea1975c67ace3a35ec9a8a40/vllm/model_executor/kernels/linear/scaled_mm/cutlass.py) is byte-identical. | +| NVFP4 | `FlashInferCuteDslNvFp4LinearKernel` | [`kernels/linear/nvfp4/flashinfer.py`](https://github.com/vllm-project/vllm/blob/dd10e03f95f94edbea1975c67ace3a35ec9a8a40/vllm/model_executor/kernels/linear/nvfp4/flashinfer.py) and the [`compressed-tensors NVFP4 scheme`](https://github.com/vllm-project/vllm/blob/dd10e03f95f94edbea1975c67ace3a35ec9a8a40/vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a4_nvfp4.py) are byte-identical. | + +This supports using the same operation, precision, shape, GPU, and kernel path +as an explicit cross-version proxy for missing measurements. It does **not** +prove identical latency across PyTorch, CUDA, FlashInfer, driver, or clock +versions. Consumers requiring only native 0.24.0 measurements must disable +shared-layer reuse; provenance identifies donor rows as `declared_reuse`. + +`test_corrected_024_gemm_uses_declared_025_measurements` compares the **entire** +native loaded GEMM map with a first-source-wins merge of all primary and donor +rows on all six systems. It catches omitted non-FP8 donor keys, unexpected extra +keys, and overwritten primary latencies. diff --git a/docs/data/pr244/case-outcomes.json.gz b/docs/data/pr244/case-outcomes.json.gz new file mode 100644 index 0000000000..7445a70af2 Binary files /dev/null and b/docs/data/pr244/case-outcomes.json.gz differ diff --git a/docs/data/pr244/manifest.json b/docs/data/pr244/manifest.json new file mode 100644 index 0000000000..50544d1972 --- /dev/null +++ b/docs/data/pr244/manifest.json @@ -0,0 +1,13866 @@ +{ + "case_outcomes_sha256": "9114213dae340e540065f7451f3a9229e197154e89832929fe7b5a23927ee9f4", + "collector_revision": "cbaf51b64fa460e5ec6146bde407a4c64958212d", + "schema_version": 1, + "source_revision": "1f29ee459882796db312b8a288c040b83edbb211", + "systems": { + "b300_sxm": { + "attempts": [ + { + "done": 37518, + "expected": 37518, + "failed": 0, + "id": "gemm-bfloat16", + "op": "gemm", + "plan_sha256": "37d5ec63f630c7a1fc241bcb3c59e2de5e52e68b54849496160dfedf0a12fe3a", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + }, + { + "done": 37518, + "expected": 37518, + "failed": 0, + "id": "gemm-fp8", + "op": "gemm", + "plan_sha256": "a26b4b0dd1684df8b9645fddcb12b080ee89d9c39f0e08b3cb44f22926a5500e", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + }, + { + "done": 37518, + "expected": 37518, + "failed": 0, + "id": "gemm-nvfp4", + "op": "gemm", + "plan_sha256": "ac15c6379b8de03e49c419d74c5b702b2596c387a07d78f833da2ff288133956", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + }, + { + "done": 300, + "expected": 300, + "failed": 0, + "id": "mla_bmm_gen_pre", + "op": "mla_bmm_gen_pre", + "plan_sha256": "3928a5f969ae055ba17244c0d53b504e8253870ffcbf60696731071ddb98e1e5", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + }, + { + "done": 336, + "expected": 336, + "failed": 0, + "id": "mla_bmm_gen_post", + "op": "mla_bmm_gen_post", + "plan_sha256": "faeaa20a84179b59df5283aa9a238572a431f4f16f8521470519147073fe4b53", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + }, + { + "done": 7679, + "expected": 7679, + "failed": 0, + "id": "encoder_attention", + "op": "encoder_attention", + "plan_sha256": "8a4f0e45a56a4b73354b445a9b3bccde722835cb6c6c133991182ba092eb1296", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + }, + { + "done": 52836, + "expected": 54468, + "failed": 1632, + "id": "attention_context", + "op": "attention_context", + "plan_sha256": "04acb1ef5df69580c3028de4dbe34f5e7697450ce783366632247395a3ed3507", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + }, + { + "done": 65408, + "expected": 67636, + "failed": 2228, + "id": "attention_generation", + "op": "attention_generation", + "plan_sha256": "12df4833c99374841693e9bec6da73b3e7a8300acb5ca88e92c5c43d4bb24cdc", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + }, + { + "done": 13059, + "expected": 13176, + "failed": 117, + "id": "mla_context_module", + "op": "mla_context_module", + "plan_sha256": "5b1dd5428859b6622f336dbc519f3c5dadc310525ee9c11d72a17557f91c8b39", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + }, + { + "done": 14137, + "expected": 17568, + "failed": 3431, + "id": "msa_context_module", + "op": "msa_context_module", + "plan_sha256": "a887b4c5331833a3ac788bdfa89e15d536beaa52bb4794a0fd27e94d05a17e47", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + }, + { + "done": 5840, + "expected": 5840, + "failed": 0, + "id": "dsv4_csa_context_module", + "op": "dsv4_csa_context_module", + "plan_sha256": "38a6cd429db806b370f13a9b847ad836440166e29d1e68c8e0c56da13396d5db", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + }, + { + "done": 9348, + "expected": 9633, + "failed": 285, + "id": "gemm-fp8_block-00", + "op": "gemm", + "plan_sha256": "e0f07c473cae107b8c2a5520c2bb1ae85045e499a268a73f0ba1ce97d035a42a", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + }, + { + "done": 9348, + "expected": 9633, + "failed": 285, + "id": "gemm-fp8_block-01", + "op": "gemm", + "plan_sha256": "fee7715156b25ee24e67995e9ae17a3bede2bbeed890c479b99cc84ade4d8fc8", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + }, + { + "done": 8856, + "expected": 9126, + "failed": 270, + "id": "gemm-fp8_block-02", + "op": "gemm", + "plan_sha256": "03318633ff86a012c7e9bf8310f97ba545972b973af130273420a0c381c85a66", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + }, + { + "done": 8856, + "expected": 9126, + "failed": 270, + "id": "gemm-fp8_block-03", + "op": "gemm", + "plan_sha256": "c5da28e5b245195dbf0816485f8230a16d02cf3aa39e8fed9dcffb8da85e87c8", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + }, + { + "done": 140, + "expected": 140, + "failed": 0, + "id": "mhc_module", + "op": "mhc_module", + "plan_sha256": "4251dd697ea2e18fcaeed92b1a7cbdf452549a75d3a38284340d5da8502e03a1", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + }, + { + "done": 3918, + "expected": 3918, + "failed": 0, + "id": "dsv4_paged_mqa_logits_module", + "op": "dsv4_paged_mqa_logits_module", + "plan_sha256": "0b35d1427f37f58a74ae69d6503f0eecb1ba5d60499325b10e1dad10a8199250", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + }, + { + "done": 3918, + "expected": 3918, + "failed": 0, + "id": "dsv4_hca_attn_module", + "op": "dsv4_hca_attn_module", + "plan_sha256": "0e844f577823301a0af954d0d20baab4642bb1c0368652998bc28160dfab816d", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + }, + { + "done": 1544, + "expected": 1544, + "failed": 0, + "id": "dsv4_csa_generation_module", + "op": "dsv4_csa_generation_module", + "plan_sha256": "21f9356813701f48d6451510032ec2746c9a64e3433c7f9954a0156c187dcae0", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + }, + { + "done": 1544, + "expected": 1544, + "failed": 0, + "id": "dsv4_hca_generation_module", + "op": "dsv4_hca_generation_module", + "plan_sha256": "f6f9fe724d2fa5ad60fb6d0a162f93cd7901bd26093925a5fef0cbae9f310eb4", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + }, + { + "done": 5840, + "expected": 5840, + "failed": 0, + "id": "dsv4_hca_context_module", + "op": "dsv4_hca_context_module", + "plan_sha256": "d36420598033f22968df6d517a051324bdac1f448ff41be9aacf12beec3dd893", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + }, + { + "done": 11400, + "expected": 13248, + "failed": 1848, + "id": "mla_generation_module", + "op": "mla_generation_module", + "plan_sha256": "3473224f524c629652b2a402f358af23f7f8a1cc2660955225ad5572182ae33e", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + }, + { + "done": 80, + "expected": 84, + "failed": 4, + "id": "gdn", + "op": "gdn", + "plan_sha256": "ba375a4c3a8dbbe8b77cec5cfd0f9dafd245e18b99ace763cc738c9645962dda", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + }, + { + "done": 2700, + "expected": 3069, + "failed": 369, + "id": "moe", + "op": "moe", + "plan_sha256": "261fd9888e2e292a179050ceb03bf8f7ab9d4f27ccd033f5075a92c165aec537", + "plan_source": "verified_checkpoint_union", + "unattempted": 0 + } + ], + "cases": { + "done": 339641, + "expected": 350380, + "failed": 10739, + "unattempted": 0 + }, + "rows": 420111, + "table_count": 17, + "tables": { + "context_attention_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "attn_dtype": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16" + ] + }, + "batch_size": { + "distinct": 9, + "max": 256, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "beam_width": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA B300 SXM6 AC" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "head_dim": { + "distinct": 5, + "max": 512, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "isl": { + "distinct": 17, + "max": 16384, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 3, + "nulls": 0, + "type": "large_string", + "values": [ + "vllm_flashinfer_flashinfertrtllmapidecode", + "vllm_flashinfer_trtllmprefill", + "vllm_triton_attn" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 29513, + "max": 767.5104166666666, + "min": 0.006709333509206772, + "nulls": 0, + "type": "double" + }, + "num_heads": { + "distinct": 17, + "max": 96, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_key_value_heads": { + "distinct": 12, + "max": 96, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "context_attention" + ] + }, + "step": { + "distinct": 1, + "max": 0, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + }, + "window_size": { + "distinct": 6, + "max": 8192, + "min": 0, + "nulls": 0, + "type": "int64" + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/attention/vllm/0.25.0/context_attention_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "batch_size", + "isl", + "num_heads", + "num_key_value_heads", + "head_dim", + "beam_width", + "attn_dtype", + "kv_cache_dtype", + "step", + "window_size" + ], + "rows": 52836, + "sha256": "776e970fb25532c0480f1bb3b666c9f34b83408bc0af7ab106bb3a01e53a1239", + "sidecar_sha256": "2dba7dd74ad73045010b63954f0f7ee668b2f7a3b4d91af527631215af6034ad" + }, + "dsv4_csa_context_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA B300 SXM6 AC" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 21, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 5254, + "max": 249.7281, + "min": 0.0576, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 5, + "max": 128, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_csa_context_module" + ] + }, + "step": { + "distinct": 24, + "max": 65535, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 4, + "max": 8, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_context_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 5840, + "sha256": "63e9b6b89418403e1998e691e2f93a5be67fd930a7997dac0d248ba0f577d7c3", + "sidecar_sha256": "bebbf880c0df4effe3b4d384e67984ddd36e254a83de2f0504e792646d6c8061" + }, + "dsv4_csa_generation_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA B300 SXM6 AC" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 854, + "max": 0.6847, + "min": 0.0583, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 5, + "max": 128, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_csa_generation_module" + ] + }, + "step": { + "distinct": 21, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 4, + "max": 8, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_generation_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 1544, + "sha256": "632a9063ce5f857ab95731449176045fbde3551efcaefb35a93feea29549e5da", + "sidecar_sha256": "bebbf880c0df4effe3b4d384e67984ddd36e254a83de2f0504e792646d6c8061" + }, + "dsv4_hca_attn_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA B300 SXM6 AC" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 16, + "max": 8192, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 1955, + "max": 5.2688, + "min": 0.0086, + "nulls": 0, + "type": "double" + }, + "local_num_heads": { + "distinct": 2, + "max": 128, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 2, + "max": 128, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_hca_attn_module" + ] + }, + "step": { + "distinct": 26, + "max": 1048575, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_attn_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "local_num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 3918, + "sha256": "cc616dbae378df19da3788d0d0b632b220144a60fb6067bbcb95cc94afe79513", + "sidecar_sha256": "bebbf880c0df4effe3b4d384e67984ddd36e254a83de2f0504e792646d6c8061" + }, + "dsv4_hca_context_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA B300 SXM6 AC" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 21, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 4914, + "max": 124.0328, + "min": 0.0414, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 5, + "max": 128, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_hca_context_module" + ] + }, + "step": { + "distinct": 24, + "max": 65535, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 4, + "max": 8, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_context_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 5840, + "sha256": "fe349c63ed6445f442f2298e47456c37ffef5f1c43d458cf5d57538d282e581e", + "sidecar_sha256": "bebbf880c0df4effe3b4d384e67984ddd36e254a83de2f0504e792646d6c8061" + }, + "dsv4_hca_generation_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA B300 SXM6 AC" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 746, + "max": 0.5444, + "min": 0.0412, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 5, + "max": 128, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_hca_generation_module" + ] + }, + "step": { + "distinct": 21, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 4, + "max": 8, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_generation_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 1544, + "sha256": "cdc48ea8a89e6b5543720df991f5f11d691b56349b1ab6a042bf6cbb550e6f62", + "sidecar_sha256": "bebbf880c0df4effe3b4d384e67984ddd36e254a83de2f0504e792646d6c8061" + }, + "dsv4_paged_mqa_logits_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA B300 SXM6 AC" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 16, + "max": 8192, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "vllm.utils.deep_gemm.fp8_fp4_paged_mqa_logits" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 769, + "max": 11.7359, + "min": 0.0044, + "nulls": 0, + "type": "double" + }, + "local_num_heads": { + "distinct": 2, + "max": 128, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_e4m3" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 2, + "max": 128, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_paged_mqa_logits_module" + ] + }, + "step": { + "distinct": 26, + "max": 1048575, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_paged_mqa_logits_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "local_num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 3918, + "sha256": "64de2420ed80948ca068da0fc1e014734e53e6a77d71ab224e566b350be2280f", + "sidecar_sha256": "bebbf880c0df4effe3b4d384e67984ddd36e254a83de2f0504e792646d6c8061" + }, + "encoder_attention_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "attn_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "batch_size": { + "distinct": 7, + "max": 64, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA B300 SXM6 AC" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "head_dim": { + "distinct": 3, + "max": 80, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "isl": { + "distinct": 39, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "vllm_vit_flash_attn_fa4" + ] + }, + "latency": { + "distinct": 5528, + "max": 63.28800964355469, + "min": 0.006506666541099548, + "nulls": 0, + "type": "double" + }, + "num_heads": { + "distinct": 12, + "max": 32, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "encoder_attention" + ] + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/encoder_attention/vllm/0.25.0/encoder_attention_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "batch_size", + "isl", + "num_heads", + "head_dim", + "attn_dtype" + ], + "rows": 7679, + "sha256": "16be054bba772c137db5f34774627410626c15eb6c3e4ba2f0d07418eddfbaac", + "sidecar_sha256": "d3b0b41bfa78a4974778e34b4a5de8f8c984e9be143b9ed1e025f849d33e7247" + }, + "gdn_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "d_conv": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "d_model": { + "distinct": 7, + "max": 8192, + "min": 1024, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA B300 SXM6 AC" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "vLLM" + ] + }, + "head_k_dim": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "head_v_dim": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 4, + "nulls": 0, + "type": "string", + "values": [ + "causal_conv1d_fn", + "causal_conv1d_update", + "chunk_gated_delta_rule_flashinfer", + "fused_recurrent_gated_delta_rule_packed_decode" + ] + }, + "latency": { + "distinct": 8914, + "max": 72.9298876953125, + "min": 0.0016918399930000306, + "nulls": 0, + "type": "double" + }, + "model_name": { + "distinct": 9, + "nulls": 0, + "type": "string", + "values": [ + "Qwen/Qwen3.5-0.8B", + "Qwen/Qwen3.5-122B-A10B", + "Qwen/Qwen3.5-27B", + "Qwen/Qwen3.5-2B", + "Qwen/Qwen3.5-35B-A3B", + "Qwen/Qwen3.5-397B-A17B", + "Qwen/Qwen3.5-4B", + "Qwen/Qwen3.5-9B", + "Qwen/Qwen3.8-2.4T-A95B" + ] + }, + "num_k_heads": { + "distinct": 5, + "max": 16, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_tokens": { + "distinct": 22, + "max": 2097152, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_v_heads": { + "distinct": 11, + "max": 128, + "min": 2, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "gdn" + ] + }, + "phase": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "context", + "generation" + ] + }, + "seq_len": { + "distinct": 16, + "max": 32768, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/linear_attention/vllm/0.25.0/gdn_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "phase", + "batch_size", + "seq_len", + "num_tokens", + "d_model", + "d_conv", + "num_k_heads", + "head_k_dim", + "num_v_heads", + "head_v_dim", + "model_name" + ], + "rows": 10326, + "sha256": "3348d7f8adb0515980553aa288473f8db9fb861b37f42e84c7a4c56024c37845", + "sidecar_sha256": "d27183136fd884307945d6de8c27e9c9fd8c11cee0fb6ca46db9bd13b74f1a0a" + }, + "gemm_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA B300 SXM6 AC" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_dtype": { + "distinct": 4, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16", + "fp8", + "fp8_block", + "nvfp4" + ] + }, + "k": { + "distinct": 23, + "max": 65536, + "min": 32, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 5, + "nulls": 0, + "type": "string", + "values": [ + "CutlassFP8ScaledMMLinearKernel", + "CutlassFp8BlockScaledMMKernel", + "DeepGemmFp8BlockScaledMMKernel", + "FlashInferCuteDslNvFp4LinearKernel", + "torch.nn.functional.linear" + ] + }, + "latency": { + "distinct": 62532, + "max": 129.59366861979166, + "min": 0.0014622221804327434, + "nulls": 0, + "type": "double" + }, + "m": { + "distinct": 74, + "max": 32768, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "n": { + "distinct": 30, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "gemm" + ] + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/gemm/vllm/0.25.0/gemm_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "gemm_dtype", + "m", + "n", + "k" + ], + "rows": 148962, + "sha256": "21328b3f7937c336452368337385f91e0e5e9f9ce067cce48a959ab9d43f014a", + "sidecar_sha256": "119d284bcbcce5ec6194e674ea51e9a0c6b7ce5759099cc4a11faa7ba8a113dd" + }, + "generation_attention_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "attn_dtype": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16" + ] + }, + "batch_size": { + "distinct": 12, + "max": 2048, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "beam_width": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA B300 SXM6 AC" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "head_dim": { + "distinct": 5, + "max": 512, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "isl": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "vllm_flashinfer_flashinfertrtllmapidecode", + "vllm_triton_attn" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 10276, + "max": 2.00652805964152, + "min": 0.006864000111818314, + "nulls": 0, + "type": "double" + }, + "num_heads": { + "distinct": 17, + "max": 96, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_key_value_heads": { + "distinct": 9, + "max": 96, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "generation_attention" + ] + }, + "step": { + "distinct": 17, + "max": 131071, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + }, + "window_size": { + "distinct": 6, + "max": 8192, + "min": 0, + "nulls": 0, + "type": "int64" + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/attention/vllm/0.25.0/generation_attention_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "batch_size", + "isl", + "num_heads", + "num_key_value_heads", + "head_dim", + "beam_width", + "attn_dtype", + "kv_cache_dtype", + "step", + "window_size" + ], + "rows": 65408, + "sha256": "74db4d37d7f75bb709e7ef24e0ea55c90138fa85a85752f7406588ffc708f2e3", + "sidecar_sha256": "2dba7dd74ad73045010b63954f0f7ee668b2f7a3b4d91af527631215af6034ad" + }, + "mhc_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA B300 SXM6 AC" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "hc_mult": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "hidden_size": { + "distinct": 2, + "max": 7168, + "min": 4096, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "vllm.model_executor.kernels.mhc.tilelang.mhc_post_tilelang", + "vllm.model_executor.kernels.mhc.tilelang.mhc_pre_tilelang" + ] + }, + "latency": { + "distinct": 112, + "max": 22.8288, + "min": 0.0065, + "nulls": 0, + "type": "double" + }, + "num_sites": { + "distinct": 1, + "max": 2, + "min": 2, + "nulls": 0, + "type": "int64" + }, + "num_tokens": { + "distinct": 35, + "max": 524288, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "post", + "pre" + ] + }, + "sinkhorn_iters": { + "distinct": 1, + "max": 20, + "min": 20, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mhc/vllm/0.25.0/mhc_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "architecture", + "num_tokens", + "num_sites", + "hc_mult", + "hidden_size", + "sinkhorn_iters" + ], + "rows": 140, + "sha256": "48480b7339c39be668c083ae42a0e2d7ef769fe6dfec453999c2ab2d33892c7e", + "sidecar_sha256": "7d57a17bd15a97a5c0f32c2f926cecb2f2aa9fc4ad063c251b4f2d68f08cfd39" + }, + "mla_bmm_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "bmm_dtype": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16" + ] + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA B300 SXM6 AC" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "vllm_torch_bmm" + ] + }, + "latency": { + "distinct": 396, + "max": 0.5412543773651123, + "min": 0.003891199827194214, + "nulls": 0, + "type": "double" + }, + "num_heads": { + "distinct": 12, + "max": 128, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_tokens": { + "distinct": 28, + "max": 20480, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "mla_gen_post", + "mla_gen_pre" + ] + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mla_bmm/vllm/0.25.0/mla_bmm_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "bmm_dtype", + "num_tokens", + "num_heads" + ], + "rows": 636, + "sha256": "1e640e27a40c8ce1e9edd7f3366f5535f39d565fd667887c98a52645cf14c90d", + "sidecar_sha256": "005809cc03ea968a721e7df4b6916b98c13458c36d77987807b36129f65db2e4" + }, + "mla_context_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "DeepseekV3ForCausalLM" + ] + }, + "batch_size": { + "distinct": 9, + "max": 256, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA B300 SXM6 AC" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 3, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8_block", + "nvfp4" + ] + }, + "isl": { + "distinct": 18, + "max": 32768, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 3, + "nulls": 0, + "type": "large_string", + "values": [ + "FLASHINFER_MLA", + "FLASH_ATTN", + "TRTLLM_RAGGED" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 8807, + "max": 115.6864, + "min": 0.0314, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "model": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "deepseek-ai/DeepSeek-V3" + ] + }, + "num_heads": { + "distinct": 12, + "max": 128, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "mla_context_module" + ] + }, + "step": { + "distinct": 1, + "max": 0, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mla/vllm/0.25.0/mla_context_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step" + ], + "rows": 13059, + "sha256": "fa7f3d1a0d1975795be09471015c7244029bfa6d56db5ad78d3bade0fc60c3fe", + "sidecar_sha256": "74f1f2cfd37deb66e5ca7c145faaf43b96cc0d0b4c23574067ea8242933bcc74" + }, + "mla_generation_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV3ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA B300 SXM6 AC" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 3, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16", + "fp8_block", + "nvfp4" + ] + }, + "isl": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHINFER_MLA" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 3460, + "max": 10.8093, + "min": 0.0313, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "deepseek-ai/DeepSeek-V3" + ] + }, + "num_heads": { + "distinct": 11, + "max": 128, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "mla_generation_module" + ] + }, + "step": { + "distinct": 17, + "max": 131072, + "min": 2, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mla/vllm/0.25.0/mla_generation_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step" + ], + "rows": 11400, + "sha256": "e563ccadb8ce7a83e4f9db434ad006b6e17ed8ae864588a56a73ccb8e060a21d", + "sidecar_sha256": "74f1f2cfd37deb66e5ca7c145faaf43b96cc0d0b4c23574067ea8242933bcc74" + }, + "moe_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA B300 SXM6 AC" + ] + }, + "distribution": { + "distinct": 3, + "nulls": 0, + "type": "large_string", + "values": [ + "balanced", + "power_law_1.01", + "power_law_1.2" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "hidden_size": { + "distinct": 12, + "max": 8192, + "min": 1024, + "nulls": 0, + "type": "int64" + }, + "inter_size": { + "distinct": 16, + "max": 16384, + "min": 512, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 14, + "nulls": 0, + "type": "large_string", + "values": [ + "vllm_compressedtensorsw4a4mxfp4moe_marlin_marlinexperts_situ_as_silu", + "vllm_compressedtensorsw4a4nvfp4moe_flashinfer_cutlass_flashinferexperts", + "vllm_compressedtensorsw4a4nvfp4moe_flashinfer_trtllm_trtllmnvfp4expertsmodular", + "vllm_compressedtensorsw4a4nvfp4moe_flashinfer_trtllm_trtllmnvfp4expertsmonolithic", + "vllm_compressedtensorswna16marlinmoe_marlin_marlinexperts", + "vllm_fp8moe_flashinfer_trtllm_trtllmfp8expertsmodular", + "vllm_fp8moe_flashinfer_trtllm_trtllmfp8expertsmonolithic", + "vllm_gptossmxfp4moe_flashinfer_trtllm_mxfp4_bf16_trtllmmxfp4expertsmonolithic", + "vllm_modeloptfp8moe_flashinfer_cutlass_flashinferexperts", + "vllm_modeloptfp8moe_flashinfer_trtllm_trtllmfp8expertsmonolithic", + "vllm_mxfp4moe_flashinfer_trtllm_mxfp4_mxfp8_trtllmmxfp4expertsmodular", + "vllm_unquantizedfusedmoe_flashinfer_cutlass_flashinferexperts", + "vllm_unquantizedfusedmoe_flashinfer_trtllm_trtllmbf16experts", + "vllm_unquantizedfusedmoe_triton_tritonexperts" + ] + }, + "latency": { + "distinct": 55664, + "max": 67.29224853515625, + "min": 0.012403199821710587, + "nulls": 0, + "type": "double" + }, + "moe_dtype": { + "distinct": 7, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8", + "fp8_block", + "int4_wo", + "nvfp4", + "w4a16_mxfp4", + "w4a8_mxfp4_mxfp8" + ] + }, + "moe_ep_size": { + "distinct": 9, + "max": 256, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "moe_tp_size": { + "distinct": 6, + "max": 32, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_experts": { + "distinct": 10, + "max": 896, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "num_tokens": { + "distinct": 27, + "max": 16384, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "moe" + ] + }, + "topk": { + "distinct": 8, + "max": 22, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/moe/vllm/0.25.0/moe_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "moe_dtype", + "num_tokens", + "hidden_size", + "inter_size", + "topk", + "num_experts", + "moe_tp_size", + "moe_ep_size", + "distribution" + ], + "rows": 72924, + "sha256": "49e0d74f2b8cb39f8f93623e2fd6445b1310c2644bb25e1a3d18703dd07e0c71", + "sidecar_sha256": "76d34fca2d48c7d0e575305d1ae2f688f697d2e6fd75d67d7cab7577cf36a378" + }, + "msa_context_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "MiniMaxM3ForCausalLM" + ] + }, + "batch_size": { + "distinct": 9, + "max": 256, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA B300 SXM6 AC" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 3, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8_block", + "nvfp4" + ] + }, + "isl": { + "distinct": 18, + "max": 32768, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "MiniMaxM3SparseMSAImpl" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 9421, + "max": 106.3639, + "min": 0.0197, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "MiniMaxAI/MiniMax-M3" + ] + }, + "num_heads": { + "distinct": 12, + "max": 128, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "msa_context_module" + ] + }, + "step": { + "distinct": 2, + "max": 128, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/msa/vllm/0.25.0/msa_context_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step" + ], + "rows": 14137, + "sha256": "ac08db7b62fa3becfa748ce943cc5e139dec54d387787236dea15e2f179f494a", + "sidecar_sha256": "824731a27227fe2c49473d07ee68c1d31ea2d48a322a39fc5d08e6faeaf94414" + } + } + }, + "gb200": { + "attempts": [ + { + "done": 37518, + "expected": 37518, + "failed": 0, + "id": "gemm-bfloat16", + "op": "gemm", + "plan_sha256": "37d5ec63f630c7a1fc241bcb3c59e2de5e52e68b54849496160dfedf0a12fe3a", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 37518, + "expected": 37518, + "failed": 0, + "id": "gemm-fp8", + "op": "gemm", + "plan_sha256": "a26b4b0dd1684df8b9645fddcb12b080ee89d9c39f0e08b3cb44f22926a5500e", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 37518, + "expected": 37518, + "failed": 0, + "id": "gemm-nvfp4", + "op": "gemm", + "plan_sha256": "ac15c6379b8de03e49c419d74c5b702b2596c387a07d78f833da2ff288133956", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 9348, + "expected": 9633, + "failed": 285, + "id": "gemm-fp8_block-00", + "op": "gemm", + "plan_sha256": "e0f07c473cae107b8c2a5520c2bb1ae85045e499a268a73f0ba1ce97d035a42a", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 9348, + "expected": 9633, + "failed": 285, + "id": "gemm-fp8_block-01", + "op": "gemm", + "plan_sha256": "fee7715156b25ee24e67995e9ae17a3bede2bbeed890c479b99cc84ade4d8fc8", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 8856, + "expected": 9126, + "failed": 270, + "id": "gemm-fp8_block-02", + "op": "gemm", + "plan_sha256": "03318633ff86a012c7e9bf8310f97ba545972b973af130273420a0c381c85a66", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 8856, + "expected": 9126, + "failed": 270, + "id": "gemm-fp8_block-03", + "op": "gemm", + "plan_sha256": "c5da28e5b245195dbf0816485f8230a16d02cf3aa39e8fed9dcffb8da85e87c8", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 52836, + "expected": 54468, + "failed": 1632, + "id": "attention_context", + "op": "attention_context", + "plan_sha256": "04acb1ef5df69580c3028de4dbe34f5e7697450ce783366632247395a3ed3507", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 13059, + "expected": 13176, + "failed": 117, + "id": "mla_context_module", + "op": "mla_context_module", + "plan_sha256": "5b1dd5428859b6622f336dbc519f3c5dadc310525ee9c11d72a17557f91c8b39", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 14108, + "expected": 17568, + "failed": 3460, + "id": "msa_context_module", + "op": "msa_context_module", + "plan_sha256": "a887b4c5331833a3ac788bdfa89e15d536beaa52bb4794a0fd27e94d05a17e47", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 2702, + "expected": 3069, + "failed": 367, + "id": "moe", + "op": "moe", + "plan_sha256": "261fd9888e2e292a179050ceb03bf8f7ab9d4f27ccd033f5075a92c165aec537", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 65408, + "expected": 67636, + "failed": 2228, + "id": "attention_generation", + "op": "attention_generation", + "plan_sha256": "12df4833c99374841693e9bec6da73b3e7a8300acb5ca88e92c5c43d4bb24cdc", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 11400, + "expected": 13248, + "failed": 1848, + "id": "mla_generation_module", + "op": "mla_generation_module", + "plan_sha256": "3473224f524c629652b2a402f358af23f7f8a1cc2660955225ad5572182ae33e", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 80, + "expected": 84, + "failed": 4, + "id": "gdn", + "op": "gdn", + "plan_sha256": "ba375a4c3a8dbbe8b77cec5cfd0f9dafd245e18b99ace763cc738c9645962dda", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 140, + "expected": 140, + "failed": 0, + "id": "mhc_module", + "op": "mhc_module", + "plan_sha256": "4251dd697ea2e18fcaeed92b1a7cbdf452549a75d3a38284340d5da8502e03a1", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 5840, + "expected": 5840, + "failed": 0, + "id": "dsv4_csa_context_module", + "op": "dsv4_csa_context_module", + "plan_sha256": "38a6cd429db806b370f13a9b847ad836440166e29d1e68c8e0c56da13396d5db", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 1544, + "expected": 1544, + "failed": 0, + "id": "dsv4_csa_generation_module", + "op": "dsv4_csa_generation_module", + "plan_sha256": "21f9356813701f48d6451510032ec2746c9a64e3433c7f9954a0156c187dcae0", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 7679, + "expected": 7679, + "failed": 0, + "id": "encoder_attention", + "op": "encoder_attention", + "plan_sha256": "8a4f0e45a56a4b73354b445a9b3bccde722835cb6c6c133991182ba092eb1296", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 300, + "expected": 300, + "failed": 0, + "id": "mla_bmm_gen_pre", + "op": "mla_bmm_gen_pre", + "plan_sha256": "3928a5f969ae055ba17244c0d53b504e8253870ffcbf60696731071ddb98e1e5", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 336, + "expected": 336, + "failed": 0, + "id": "mla_bmm_gen_post", + "op": "mla_bmm_gen_post", + "plan_sha256": "faeaa20a84179b59df5283aa9a238572a431f4f16f8521470519147073fe4b53", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 5840, + "expected": 5840, + "failed": 0, + "id": "dsv4_hca_context_module", + "op": "dsv4_hca_context_module", + "plan_sha256": "d36420598033f22968df6d517a051324bdac1f448ff41be9aacf12beec3dd893", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 1544, + "expected": 1544, + "failed": 0, + "id": "dsv4_hca_generation_module", + "op": "dsv4_hca_generation_module", + "plan_sha256": "f6f9fe724d2fa5ad60fb6d0a162f93cd7901bd26093925a5fef0cbae9f310eb4", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 3918, + "expected": 3918, + "failed": 0, + "id": "dsv4_paged_mqa_logits_module", + "op": "dsv4_paged_mqa_logits_module", + "plan_sha256": "0b35d1427f37f58a74ae69d6503f0eecb1ba5d60499325b10e1dad10a8199250", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 3918, + "expected": 3918, + "failed": 0, + "id": "dsv4_hca_attn_module", + "op": "dsv4_hca_attn_module", + "plan_sha256": "0e844f577823301a0af954d0d20baab4642bb1c0368652998bc28160dfab816d", + "plan_source": "archived_case_plan", + "unattempted": 0 + } + ], + "cases": { + "done": 339614, + "expected": 350380, + "failed": 10766, + "unattempted": 0 + }, + "rows": 420134, + "table_count": 17, + "tables": { + "context_attention_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "attn_dtype": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16" + ] + }, + "batch_size": { + "distinct": 9, + "max": 256, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "beam_width": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA GB200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "head_dim": { + "distinct": 5, + "max": 512, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "isl": { + "distinct": 17, + "max": 16384, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 3, + "nulls": 0, + "type": "large_string", + "values": [ + "vllm_flashinfer_flashinfertrtllmapidecode", + "vllm_flashinfer_trtllmprefill", + "vllm_triton_attn" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 29848, + "max": 735.9351399739584, + "min": 0.008634666601816813, + "nulls": 0, + "type": "double" + }, + "num_heads": { + "distinct": 17, + "max": 96, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_key_value_heads": { + "distinct": 12, + "max": 96, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "context_attention" + ] + }, + "step": { + "distinct": 1, + "max": 0, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + }, + "window_size": { + "distinct": 6, + "max": 8192, + "min": 0, + "nulls": 0, + "type": "int64" + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb200/attention/vllm/0.25.0/context_attention_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "batch_size", + "isl", + "num_heads", + "num_key_value_heads", + "head_dim", + "beam_width", + "attn_dtype", + "kv_cache_dtype", + "step", + "window_size" + ], + "rows": 52836, + "sha256": "cda22b9ab7d5f1d15e6b8b3cbe6660bdce0dbb6402abff781a8f77a70a82209f", + "sidecar_sha256": "a5794ac332016a0982121fb23cc94fb05350f154d1a8770f11569b23c8b6ca24" + }, + "dsv4_csa_context_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA GB200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 21, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 5279, + "max": 242.1415, + "min": 0.0539, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 5, + "max": 128, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_csa_context_module" + ] + }, + "step": { + "distinct": 24, + "max": 65535, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 4, + "max": 8, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_csa_context_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 5840, + "sha256": "76719d88e0481b5153c89ef0e1024435597f660de305bdb9f01750e9273189d3", + "sidecar_sha256": "ff4b25e84cb1c03a7c2b6bec9adc91e731a2dd57bb96d72089e13a4440d0d674" + }, + "dsv4_csa_generation_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA GB200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 886, + "max": 0.6834, + "min": 0.0543, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 5, + "max": 128, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_csa_generation_module" + ] + }, + "step": { + "distinct": 21, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 4, + "max": 8, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_csa_generation_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 1544, + "sha256": "b1a09a8e004ebed907e1d6d85379c51f5ec4ca2c5812bb59a62327c99f2dc505", + "sidecar_sha256": "ff4b25e84cb1c03a7c2b6bec9adc91e731a2dd57bb96d72089e13a4440d0d674" + }, + "dsv4_hca_attn_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA GB200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 16, + "max": 8192, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 1985, + "max": 5.0301, + "min": 0.0106, + "nulls": 0, + "type": "double" + }, + "local_num_heads": { + "distinct": 2, + "max": 128, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 2, + "max": 128, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_hca_attn_module" + ] + }, + "step": { + "distinct": 26, + "max": 1048575, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_hca_attn_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "local_num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 3918, + "sha256": "0abf339525484c011c1c57508bbf01d1338f98f93cbc3d0debc6b2b383b2d0dc", + "sidecar_sha256": "ff4b25e84cb1c03a7c2b6bec9adc91e731a2dd57bb96d72089e13a4440d0d674" + }, + "dsv4_hca_context_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA GB200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 21, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 4931, + "max": 120.7941, + "min": 0.0436, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 5, + "max": 128, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_hca_context_module" + ] + }, + "step": { + "distinct": 24, + "max": 65535, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 4, + "max": 8, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_hca_context_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 5840, + "sha256": "633e6a30fc69935c0edd9fb672d26a02ba3d13a36e5412f5dab57571b37c7ec1", + "sidecar_sha256": "ff4b25e84cb1c03a7c2b6bec9adc91e731a2dd57bb96d72089e13a4440d0d674" + }, + "dsv4_hca_generation_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA GB200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 740, + "max": 0.5519, + "min": 0.0421, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 5, + "max": 128, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_hca_generation_module" + ] + }, + "step": { + "distinct": 21, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 4, + "max": 8, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_hca_generation_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 1544, + "sha256": "9807d59d5b0e6a14a5d82feb7b81c8529fbe2d0d65120fd32706a7b68fe4cf2a", + "sidecar_sha256": "ff4b25e84cb1c03a7c2b6bec9adc91e731a2dd57bb96d72089e13a4440d0d674" + }, + "dsv4_paged_mqa_logits_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA GB200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 16, + "max": 8192, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "vllm.utils.deep_gemm.fp8_fp4_paged_mqa_logits" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 992, + "max": 11.271, + "min": 0.0062, + "nulls": 0, + "type": "double" + }, + "local_num_heads": { + "distinct": 2, + "max": 128, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_e4m3" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 2, + "max": 128, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_paged_mqa_logits_module" + ] + }, + "step": { + "distinct": 26, + "max": 1048575, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_paged_mqa_logits_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "local_num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 3918, + "sha256": "e62d12cc3c8a125b66ad066de2abfda101c06a6acdbf9f7d0e9bebd4f442f4c3", + "sidecar_sha256": "ff4b25e84cb1c03a7c2b6bec9adc91e731a2dd57bb96d72089e13a4440d0d674" + }, + "encoder_attention_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "attn_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "batch_size": { + "distinct": 7, + "max": 64, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA GB200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "head_dim": { + "distinct": 3, + "max": 80, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "isl": { + "distinct": 39, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "vllm_vit_flash_attn_fa4" + ] + }, + "latency": { + "distinct": 5670, + "max": 67.4124247233073, + "min": 0.006655999769767125, + "nulls": 0, + "type": "double" + }, + "num_heads": { + "distinct": 12, + "max": 32, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "encoder_attention" + ] + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb200/encoder_attention/vllm/0.25.0/encoder_attention_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "batch_size", + "isl", + "num_heads", + "head_dim", + "attn_dtype" + ], + "rows": 7679, + "sha256": "22bf62b5de4b28154065531477f1b902469a9700ac5ab783df0de23f0706837a", + "sidecar_sha256": "76cb9c54b63248b53f5c43643f0fa367300f61f76f2122baae34998903cfd715" + }, + "gdn_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "d_conv": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "d_model": { + "distinct": 7, + "max": 8192, + "min": 1024, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA GB200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "vLLM" + ] + }, + "head_k_dim": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "head_v_dim": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 4, + "nulls": 0, + "type": "string", + "values": [ + "causal_conv1d_fn", + "causal_conv1d_update", + "chunk_gated_delta_rule_flashinfer", + "fused_recurrent_gated_delta_rule_packed_decode" + ] + }, + "latency": { + "distinct": 9261, + "max": 70.44921875, + "min": 0.0019068799912929535, + "nulls": 0, + "type": "double" + }, + "model_name": { + "distinct": 9, + "nulls": 0, + "type": "string", + "values": [ + "Qwen/Qwen3.5-0.8B", + "Qwen/Qwen3.5-122B-A10B", + "Qwen/Qwen3.5-27B", + "Qwen/Qwen3.5-2B", + "Qwen/Qwen3.5-35B-A3B", + "Qwen/Qwen3.5-397B-A17B", + "Qwen/Qwen3.5-4B", + "Qwen/Qwen3.5-9B", + "Qwen/Qwen3.8-2.4T-A95B" + ] + }, + "num_k_heads": { + "distinct": 5, + "max": 16, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_tokens": { + "distinct": 22, + "max": 2097152, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_v_heads": { + "distinct": 11, + "max": 128, + "min": 2, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "gdn" + ] + }, + "phase": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "context", + "generation" + ] + }, + "seq_len": { + "distinct": 16, + "max": 32768, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb200/linear_attention/vllm/0.25.0/gdn_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "phase", + "batch_size", + "seq_len", + "num_tokens", + "d_model", + "d_conv", + "num_k_heads", + "head_k_dim", + "num_v_heads", + "head_v_dim", + "model_name" + ], + "rows": 10326, + "sha256": "6cf42886c237ce26c7c79ad9b7cc8bbeac9dd6c84c657221c1713398907fe6b9", + "sidecar_sha256": "d7e55c45e71f01589273aabebfe23a3e67d5a6bed9111ef2b11fb74bb8848352" + }, + "gemm_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA GB200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_dtype": { + "distinct": 4, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16", + "fp8", + "fp8_block", + "nvfp4" + ] + }, + "k": { + "distinct": 23, + "max": 65536, + "min": 32, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 5, + "nulls": 0, + "type": "string", + "values": [ + "CutlassFP8ScaledMMLinearKernel", + "CutlassFp8BlockScaledMMKernel", + "DeepGemmFp8BlockScaledMMKernel", + "FlashInferCuteDslNvFp4LinearKernel", + "torch.nn.functional.linear" + ] + }, + "latency": { + "distinct": 62254, + "max": 120.95967610677083, + "min": 0.001769777801301744, + "nulls": 0, + "type": "double" + }, + "m": { + "distinct": 74, + "max": 32768, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "n": { + "distinct": 30, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "gemm" + ] + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb200/gemm/vllm/0.25.0/gemm_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "gemm_dtype", + "m", + "n", + "k" + ], + "rows": 148962, + "sha256": "8aad335ba9b743f50f9c168eb085106c06ac42a06c3f6c4e415c856ced5fb21d", + "sidecar_sha256": "1ab1f48faf214cb9be95ee5b6a3f92403d13c03b6ae32f82aa3b389a80e2b79e" + }, + "generation_attention_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "attn_dtype": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16" + ] + }, + "batch_size": { + "distinct": 12, + "max": 2048, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "beam_width": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA GB200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "head_dim": { + "distinct": 5, + "max": 512, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "isl": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "vllm_flashinfer_flashinfertrtllmapidecode", + "vllm_triton_attn" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 10174, + "max": 1.9459946950276692, + "min": 0.008506666868925095, + "nulls": 0, + "type": "double" + }, + "num_heads": { + "distinct": 17, + "max": 96, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_key_value_heads": { + "distinct": 9, + "max": 96, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "generation_attention" + ] + }, + "step": { + "distinct": 17, + "max": 131071, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + }, + "window_size": { + "distinct": 6, + "max": 8192, + "min": 0, + "nulls": 0, + "type": "int64" + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb200/attention/vllm/0.25.0/generation_attention_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "batch_size", + "isl", + "num_heads", + "num_key_value_heads", + "head_dim", + "beam_width", + "attn_dtype", + "kv_cache_dtype", + "step", + "window_size" + ], + "rows": 65408, + "sha256": "bc41206e6d2411149ffecb7d86ba0e1cb86f4a9426d6b5b2700fb457b3367cf9", + "sidecar_sha256": "a5794ac332016a0982121fb23cc94fb05350f154d1a8770f11569b23c8b6ca24" + }, + "mhc_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA GB200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "hc_mult": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "hidden_size": { + "distinct": 2, + "max": 7168, + "min": 4096, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "vllm.model_executor.kernels.mhc.tilelang.mhc_post_tilelang", + "vllm.model_executor.kernels.mhc.tilelang.mhc_pre_tilelang" + ] + }, + "latency": { + "distinct": 104, + "max": 22.05, + "min": 0.0085, + "nulls": 0, + "type": "double" + }, + "num_sites": { + "distinct": 1, + "max": 2, + "min": 2, + "nulls": 0, + "type": "int64" + }, + "num_tokens": { + "distinct": 35, + "max": 524288, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "post", + "pre" + ] + }, + "sinkhorn_iters": { + "distinct": 1, + "max": 20, + "min": 20, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mhc/vllm/0.25.0/mhc_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "architecture", + "num_tokens", + "num_sites", + "hc_mult", + "hidden_size", + "sinkhorn_iters" + ], + "rows": 140, + "sha256": "ed07da30bb350d5999a3a974beb3aa70c2282fed748ce60c197352c8951d3cd8", + "sidecar_sha256": "aa54deb2dbdf3173fe328db70ec565b1eea966ba9d488c22bb825d556d3e85c2" + }, + "mla_bmm_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "bmm_dtype": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16" + ] + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA GB200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "vllm_torch_bmm" + ] + }, + "latency": { + "distinct": 489, + "max": 0.5524032115936279, + "min": 0.0044096000492572784, + "nulls": 0, + "type": "double" + }, + "num_heads": { + "distinct": 12, + "max": 128, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_tokens": { + "distinct": 28, + "max": 20480, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "mla_gen_post", + "mla_gen_pre" + ] + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mla_bmm/vllm/0.25.0/mla_bmm_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "bmm_dtype", + "num_tokens", + "num_heads" + ], + "rows": 636, + "sha256": "7a769fafe155446dbd8054b16c81a27b0ca892a75d4a1c673c4592c853cc284d", + "sidecar_sha256": "865749cd01284bdee328aad42bf7bf70d958bc99b0cee8916191203686b195d1" + }, + "mla_context_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "DeepseekV3ForCausalLM" + ] + }, + "batch_size": { + "distinct": 9, + "max": 256, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA GB200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 3, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8_block", + "nvfp4" + ] + }, + "isl": { + "distinct": 18, + "max": 32768, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 3, + "nulls": 0, + "type": "large_string", + "values": [ + "FLASHINFER_MLA", + "FLASH_ATTN", + "TRTLLM_RAGGED" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 8797, + "max": 112.3707, + "min": 0.0324, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "model": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "deepseek-ai/DeepSeek-V3" + ] + }, + "num_heads": { + "distinct": 12, + "max": 128, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "mla_context_module" + ] + }, + "step": { + "distinct": 1, + "max": 0, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mla/vllm/0.25.0/mla_context_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step" + ], + "rows": 13059, + "sha256": "cece8d599135add23eb20f7fdd8465f58e9f6b9f1831846990c7be37c03f3a9a", + "sidecar_sha256": "df340d8b0c11b8a7b6abbe52054312e24990c9580b34dca78dca7615809c2234" + }, + "mla_generation_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV3ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA GB200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 3, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16", + "fp8_block", + "nvfp4" + ] + }, + "isl": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHINFER_MLA" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 3456, + "max": 10.5559, + "min": 0.0318, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "deepseek-ai/DeepSeek-V3" + ] + }, + "num_heads": { + "distinct": 11, + "max": 128, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "mla_generation_module" + ] + }, + "step": { + "distinct": 17, + "max": 131072, + "min": 2, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mla/vllm/0.25.0/mla_generation_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step" + ], + "rows": 11400, + "sha256": "be60482ec3dabe154b07b42e3a3fa809f2a93cf6e031b7b22202c92d5b172ab4", + "sidecar_sha256": "df340d8b0c11b8a7b6abbe52054312e24990c9580b34dca78dca7615809c2234" + }, + "moe_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA GB200" + ] + }, + "distribution": { + "distinct": 3, + "nulls": 0, + "type": "large_string", + "values": [ + "balanced", + "power_law_1.01", + "power_law_1.2" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "hidden_size": { + "distinct": 12, + "max": 8192, + "min": 1024, + "nulls": 0, + "type": "int64" + }, + "inter_size": { + "distinct": 16, + "max": 16384, + "min": 512, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 14, + "nulls": 0, + "type": "large_string", + "values": [ + "vllm_compressedtensorsw4a4mxfp4moe_marlin_marlinexperts_situ_as_silu", + "vllm_compressedtensorsw4a4nvfp4moe_flashinfer_cutlass_flashinferexperts", + "vllm_compressedtensorsw4a4nvfp4moe_flashinfer_trtllm_trtllmnvfp4expertsmodular", + "vllm_compressedtensorsw4a4nvfp4moe_flashinfer_trtllm_trtllmnvfp4expertsmonolithic", + "vllm_compressedtensorswna16marlinmoe_marlin_marlinexperts", + "vllm_fp8moe_flashinfer_trtllm_trtllmfp8expertsmodular", + "vllm_fp8moe_flashinfer_trtllm_trtllmfp8expertsmonolithic", + "vllm_gptossmxfp4moe_flashinfer_trtllm_mxfp4_bf16_trtllmmxfp4expertsmonolithic", + "vllm_modeloptfp8moe_flashinfer_cutlass_flashinferexperts", + "vllm_modeloptfp8moe_flashinfer_trtllm_trtllmfp8expertsmonolithic", + "vllm_mxfp4moe_flashinfer_trtllm_mxfp4_mxfp8_trtllmmxfp4expertsmodular", + "vllm_unquantizedfusedmoe_flashinfer_cutlass_flashinferexperts", + "vllm_unquantizedfusedmoe_flashinfer_trtllm_trtllmbf16experts", + "vllm_unquantizedfusedmoe_triton_tritonexperts" + ] + }, + "latency": { + "distinct": 55416, + "max": 64.66356201171875, + "min": 0.013068799674510957, + "nulls": 0, + "type": "double" + }, + "moe_dtype": { + "distinct": 7, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8", + "fp8_block", + "int4_wo", + "nvfp4", + "w4a16_mxfp4", + "w4a8_mxfp4_mxfp8" + ] + }, + "moe_ep_size": { + "distinct": 9, + "max": 256, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "moe_tp_size": { + "distinct": 6, + "max": 32, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_experts": { + "distinct": 10, + "max": 896, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "num_tokens": { + "distinct": 27, + "max": 16384, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "moe" + ] + }, + "topk": { + "distinct": 8, + "max": 22, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb200/moe/vllm/0.25.0/moe_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "moe_dtype", + "num_tokens", + "hidden_size", + "inter_size", + "topk", + "num_experts", + "moe_tp_size", + "moe_ep_size", + "distribution" + ], + "rows": 72976, + "sha256": "619956a1ef46af90ff89e4e7ec742f230bede14b145f74ccb6ee53fde60a6265", + "sidecar_sha256": "55b0afa738bbca95798e32ef6432c8f9fb83543e722ad658e476b7bf29f2a622" + }, + "msa_context_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "MiniMaxM3ForCausalLM" + ] + }, + "batch_size": { + "distinct": 9, + "max": 256, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA GB200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 3, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8_block", + "nvfp4" + ] + }, + "isl": { + "distinct": 18, + "max": 32768, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "MiniMaxM3SparseMSAImpl" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 9537, + "max": 102.6411, + "min": 0.021, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "MiniMaxAI/MiniMax-M3" + ] + }, + "num_heads": { + "distinct": 12, + "max": 128, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "msa_context_module" + ] + }, + "step": { + "distinct": 2, + "max": 128, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb200/msa/vllm/0.25.0/msa_context_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step" + ], + "rows": 14108, + "sha256": "38967e8b8ac41c19efceb17674d71de6e30b6d9f7ca17e77c2c983dd2ad8cb23", + "sidecar_sha256": "fea11c03782b330c875c36500130d2e94cdb202b35a4fa1a1cf7e9508bd6b561" + } + } + }, + "gb300": { + "attempts": [ + { + "done": 37518, + "expected": 37518, + "failed": 0, + "id": "gemm-bfloat16", + "op": "gemm", + "plan_sha256": "37d5ec63f630c7a1fc241bcb3c59e2de5e52e68b54849496160dfedf0a12fe3a", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 37518, + "expected": 37518, + "failed": 0, + "id": "gemm-fp8", + "op": "gemm", + "plan_sha256": "a26b4b0dd1684df8b9645fddcb12b080ee89d9c39f0e08b3cb44f22926a5500e", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 37518, + "expected": 37518, + "failed": 0, + "id": "gemm-nvfp4", + "op": "gemm", + "plan_sha256": "ac15c6379b8de03e49c419d74c5b702b2596c387a07d78f833da2ff288133956", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 52836, + "expected": 54468, + "failed": 1632, + "id": "attention_context", + "op": "attention_context", + "plan_sha256": "04acb1ef5df69580c3028de4dbe34f5e7697450ce783366632247395a3ed3507", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 13059, + "expected": 13176, + "failed": 117, + "id": "mla_context_module", + "op": "mla_context_module", + "plan_sha256": "5b1dd5428859b6622f336dbc519f3c5dadc310525ee9c11d72a17557f91c8b39", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 65408, + "expected": 67636, + "failed": 2228, + "id": "attention_generation", + "op": "attention_generation", + "plan_sha256": "12df4833c99374841693e9bec6da73b3e7a8300acb5ca88e92c5c43d4bb24cdc", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 11400, + "expected": 13248, + "failed": 1848, + "id": "mla_generation_module", + "op": "mla_generation_module", + "plan_sha256": "3473224f524c629652b2a402f358af23f7f8a1cc2660955225ad5572182ae33e", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 7679, + "expected": 7679, + "failed": 0, + "id": "encoder_attention", + "op": "encoder_attention", + "plan_sha256": "8a4f0e45a56a4b73354b445a9b3bccde722835cb6c6c133991182ba092eb1296", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 300, + "expected": 300, + "failed": 0, + "id": "mla_bmm_gen_pre", + "op": "mla_bmm_gen_pre", + "plan_sha256": "3928a5f969ae055ba17244c0d53b504e8253870ffcbf60696731071ddb98e1e5", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 336, + "expected": 336, + "failed": 0, + "id": "mla_bmm_gen_post", + "op": "mla_bmm_gen_post", + "plan_sha256": "faeaa20a84179b59df5283aa9a238572a431f4f16f8521470519147073fe4b53", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 80, + "expected": 84, + "failed": 4, + "id": "gdn", + "op": "gdn", + "plan_sha256": "ba375a4c3a8dbbe8b77cec5cfd0f9dafd245e18b99ace763cc738c9645962dda", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 140, + "expected": 140, + "failed": 0, + "id": "mhc_module", + "op": "mhc_module", + "plan_sha256": "4251dd697ea2e18fcaeed92b1a7cbdf452549a75d3a38284340d5da8502e03a1", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 5840, + "expected": 5840, + "failed": 0, + "id": "dsv4_csa_context_module", + "op": "dsv4_csa_context_module", + "plan_sha256": "38a6cd429db806b370f13a9b847ad836440166e29d1e68c8e0c56da13396d5db", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 1544, + "expected": 1544, + "failed": 0, + "id": "dsv4_csa_generation_module", + "op": "dsv4_csa_generation_module", + "plan_sha256": "21f9356813701f48d6451510032ec2746c9a64e3433c7f9954a0156c187dcae0", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 5840, + "expected": 5840, + "failed": 0, + "id": "dsv4_hca_context_module", + "op": "dsv4_hca_context_module", + "plan_sha256": "d36420598033f22968df6d517a051324bdac1f448ff41be9aacf12beec3dd893", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 1544, + "expected": 1544, + "failed": 0, + "id": "dsv4_hca_generation_module", + "op": "dsv4_hca_generation_module", + "plan_sha256": "f6f9fe724d2fa5ad60fb6d0a162f93cd7901bd26093925a5fef0cbae9f310eb4", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 3918, + "expected": 3918, + "failed": 0, + "id": "dsv4_paged_mqa_logits_module", + "op": "dsv4_paged_mqa_logits_module", + "plan_sha256": "0b35d1427f37f58a74ae69d6503f0eecb1ba5d60499325b10e1dad10a8199250", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 3918, + "expected": 3918, + "failed": 0, + "id": "dsv4_hca_attn_module", + "op": "dsv4_hca_attn_module", + "plan_sha256": "0e844f577823301a0af954d0d20baab4642bb1c0368652998bc28160dfab816d", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 9348, + "expected": 9633, + "failed": 285, + "id": "gemm-fp8_block-00", + "op": "gemm", + "plan_sha256": "e0f07c473cae107b8c2a5520c2bb1ae85045e499a268a73f0ba1ce97d035a42a", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 9348, + "expected": 9633, + "failed": 285, + "id": "gemm-fp8_block-01", + "op": "gemm", + "plan_sha256": "fee7715156b25ee24e67995e9ae17a3bede2bbeed890c479b99cc84ade4d8fc8", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 8856, + "expected": 9126, + "failed": 270, + "id": "gemm-fp8_block-02", + "op": "gemm", + "plan_sha256": "03318633ff86a012c7e9bf8310f97ba545972b973af130273420a0c381c85a66", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 8856, + "expected": 9126, + "failed": 270, + "id": "gemm-fp8_block-03", + "op": "gemm", + "plan_sha256": "c5da28e5b245195dbf0816485f8230a16d02cf3aa39e8fed9dcffb8da85e87c8", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 14174, + "expected": 17568, + "failed": 3394, + "id": "msa_context_module", + "op": "msa_context_module", + "plan_sha256": "a887b4c5331833a3ac788bdfa89e15d536beaa52bb4794a0fd27e94d05a17e47", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 2703, + "expected": 3069, + "failed": 366, + "id": "moe", + "op": "moe", + "plan_sha256": "261fd9888e2e292a179050ceb03bf8f7ab9d4f27ccd033f5075a92c165aec537", + "plan_source": "archived_case_plan", + "unattempted": 0 + } + ], + "cases": { + "done": 339681, + "expected": 350380, + "failed": 10699, + "unattempted": 0 + }, + "rows": 420226, + "table_count": 17, + "tables": { + "context_attention_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "attn_dtype": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16" + ] + }, + "batch_size": { + "distinct": 9, + "max": 256, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "beam_width": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA GB300" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "head_dim": { + "distinct": 5, + "max": 512, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "isl": { + "distinct": 17, + "max": 16384, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 3, + "nulls": 0, + "type": "large_string", + "values": [ + "vllm_flashinfer_flashinfertrtllmapidecode", + "vllm_flashinfer_trtllmprefill", + "vllm_triton_attn" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 29330, + "max": 727.7124837239584, + "min": 0.008576000109314919, + "nulls": 0, + "type": "double" + }, + "num_heads": { + "distinct": 17, + "max": 96, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_key_value_heads": { + "distinct": 12, + "max": 96, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "context_attention" + ] + }, + "step": { + "distinct": 1, + "max": 0, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + }, + "window_size": { + "distinct": 6, + "max": 8192, + "min": 0, + "nulls": 0, + "type": "int64" + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb300/attention/vllm/0.25.0/context_attention_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "batch_size", + "isl", + "num_heads", + "num_key_value_heads", + "head_dim", + "beam_width", + "attn_dtype", + "kv_cache_dtype", + "step", + "window_size" + ], + "rows": 52836, + "sha256": "1e6e062bff48e40ea5db33fc95b095c3b23c967a6e94a32a197a7d53adae74c9", + "sidecar_sha256": "a5794ac332016a0982121fb23cc94fb05350f154d1a8770f11569b23c8b6ca24" + }, + "dsv4_csa_context_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA GB300" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 21, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 5259, + "max": 239.2823, + "min": 0.0526, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 5, + "max": 128, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_csa_context_module" + ] + }, + "step": { + "distinct": 24, + "max": 65535, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 4, + "max": 8, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_csa_context_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 5840, + "sha256": "81276eea4e424b2146b58256cd1697529a3a00c8bf3ee7c6613d54d6094972fc", + "sidecar_sha256": "ff4b25e84cb1c03a7c2b6bec9adc91e731a2dd57bb96d72089e13a4440d0d674" + }, + "dsv4_csa_generation_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA GB300" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 870, + "max": 0.6709, + "min": 0.0533, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 5, + "max": 128, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_csa_generation_module" + ] + }, + "step": { + "distinct": 21, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 4, + "max": 8, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_csa_generation_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 1544, + "sha256": "eb54cfe271dc3d15d0849745c17cb1979da4ac4afc63504acece52bcf4482d08", + "sidecar_sha256": "ff4b25e84cb1c03a7c2b6bec9adc91e731a2dd57bb96d72089e13a4440d0d674" + }, + "dsv4_hca_attn_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA GB300" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 16, + "max": 8192, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 1920, + "max": 4.9961, + "min": 0.0106, + "nulls": 0, + "type": "double" + }, + "local_num_heads": { + "distinct": 2, + "max": 128, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 2, + "max": 128, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_hca_attn_module" + ] + }, + "step": { + "distinct": 26, + "max": 1048575, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_hca_attn_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "local_num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 3918, + "sha256": "49ecc0d7e34e1503f7bdbea1c3599bb20a219488786a22fef43cf216de641383", + "sidecar_sha256": "ff4b25e84cb1c03a7c2b6bec9adc91e731a2dd57bb96d72089e13a4440d0d674" + }, + "dsv4_hca_context_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA GB300" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 21, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 4935, + "max": 118.5108, + "min": 0.0435, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 5, + "max": 128, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_hca_context_module" + ] + }, + "step": { + "distinct": 24, + "max": 65535, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 4, + "max": 8, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_hca_context_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 5840, + "sha256": "56d026f2e2b5b843c0afbe6b6c0d9b3489abf5db256d5d61ee065f7bcd72d15f", + "sidecar_sha256": "ff4b25e84cb1c03a7c2b6bec9adc91e731a2dd57bb96d72089e13a4440d0d674" + }, + "dsv4_hca_generation_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA GB300" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 712, + "max": 0.5354, + "min": 0.0416, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 5, + "max": 128, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_hca_generation_module" + ] + }, + "step": { + "distinct": 21, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 4, + "max": 8, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_hca_generation_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 1544, + "sha256": "3dee3c879fe7fdd9fd1e111d2f8df79cf15d8e635ebe32f123b94adc441fc31e", + "sidecar_sha256": "ff4b25e84cb1c03a7c2b6bec9adc91e731a2dd57bb96d72089e13a4440d0d674" + }, + "dsv4_paged_mqa_logits_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA GB300" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 16, + "max": 8192, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "vllm.utils.deep_gemm.fp8_fp4_paged_mqa_logits" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 887, + "max": 10.3622, + "min": 0.0063, + "nulls": 0, + "type": "double" + }, + "local_num_heads": { + "distinct": 2, + "max": 128, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_e4m3" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 2, + "max": 128, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_paged_mqa_logits_module" + ] + }, + "step": { + "distinct": 26, + "max": 1048575, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_paged_mqa_logits_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "local_num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 3918, + "sha256": "12f3aa8b290193bf97e2f144c9bdda347eeba8ab10698669bc41d1fa617280fd", + "sidecar_sha256": "ff4b25e84cb1c03a7c2b6bec9adc91e731a2dd57bb96d72089e13a4440d0d674" + }, + "encoder_attention_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "attn_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "batch_size": { + "distinct": 7, + "max": 64, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA GB300" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "head_dim": { + "distinct": 3, + "max": 80, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "isl": { + "distinct": 39, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "vllm_vit_flash_attn_fa4" + ] + }, + "latency": { + "distinct": 5602, + "max": 54.99699401855469, + "min": 0.006624000146985054, + "nulls": 0, + "type": "double" + }, + "num_heads": { + "distinct": 12, + "max": 32, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "encoder_attention" + ] + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb300/encoder_attention/vllm/0.25.0/encoder_attention_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "batch_size", + "isl", + "num_heads", + "head_dim", + "attn_dtype" + ], + "rows": 7679, + "sha256": "7b90118f5ecc35f18087c127eecd4808360c8c5e2bde26a2ff391304b304c303", + "sidecar_sha256": "76cb9c54b63248b53f5c43643f0fa367300f61f76f2122baae34998903cfd715" + }, + "gdn_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "d_conv": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "d_model": { + "distinct": 7, + "max": 8192, + "min": 1024, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA GB300" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "vLLM" + ] + }, + "head_k_dim": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "head_v_dim": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 4, + "nulls": 0, + "type": "string", + "values": [ + "causal_conv1d_fn", + "causal_conv1d_update", + "chunk_gated_delta_rule_flashinfer", + "fused_recurrent_gated_delta_rule_packed_decode" + ] + }, + "latency": { + "distinct": 9448, + "max": 66.970615234375, + "min": 0.0018860800564289091, + "nulls": 0, + "type": "double" + }, + "model_name": { + "distinct": 9, + "nulls": 0, + "type": "string", + "values": [ + "Qwen/Qwen3.5-0.8B", + "Qwen/Qwen3.5-122B-A10B", + "Qwen/Qwen3.5-27B", + "Qwen/Qwen3.5-2B", + "Qwen/Qwen3.5-35B-A3B", + "Qwen/Qwen3.5-397B-A17B", + "Qwen/Qwen3.5-4B", + "Qwen/Qwen3.5-9B", + "Qwen/Qwen3.8-2.4T-A95B" + ] + }, + "num_k_heads": { + "distinct": 5, + "max": 16, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_tokens": { + "distinct": 22, + "max": 2097152, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_v_heads": { + "distinct": 11, + "max": 128, + "min": 2, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "gdn" + ] + }, + "phase": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "context", + "generation" + ] + }, + "seq_len": { + "distinct": 16, + "max": 32768, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb300/linear_attention/vllm/0.25.0/gdn_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "phase", + "batch_size", + "seq_len", + "num_tokens", + "d_model", + "d_conv", + "num_k_heads", + "head_k_dim", + "num_v_heads", + "head_v_dim", + "model_name" + ], + "rows": 10326, + "sha256": "c4d76ff304ba8e098551ac9d982d8dbc1746effd07ee0b5a0ce67a6e4cebc955", + "sidecar_sha256": "d7e55c45e71f01589273aabebfe23a3e67d5a6bed9111ef2b11fb74bb8848352" + }, + "gemm_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA GB300" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_dtype": { + "distinct": 4, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16", + "fp8", + "fp8_block", + "nvfp4" + ] + }, + "k": { + "distinct": 23, + "max": 65536, + "min": 32, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 5, + "nulls": 0, + "type": "string", + "values": [ + "CutlassFP8ScaledMMLinearKernel", + "CutlassFp8BlockScaledMMKernel", + "DeepGemmFp8BlockScaledMMKernel", + "FlashInferCuteDslNvFp4LinearKernel", + "torch.nn.functional.linear" + ] + }, + "latency": { + "distinct": 61500, + "max": 103.81315782335069, + "min": 0.0017866666118303935, + "nulls": 0, + "type": "double" + }, + "m": { + "distinct": 74, + "max": 32768, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "n": { + "distinct": 30, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "gemm" + ] + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb300/gemm/vllm/0.25.0/gemm_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "gemm_dtype", + "m", + "n", + "k" + ], + "rows": 148962, + "sha256": "ee2b29e554798c581b3d7e574f354425ea5264719758ddc97c369db37f5972f3", + "sidecar_sha256": "1ab1f48faf214cb9be95ee5b6a3f92403d13c03b6ae32f82aa3b389a80e2b79e" + }, + "generation_attention_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "attn_dtype": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16" + ] + }, + "batch_size": { + "distinct": 12, + "max": 2048, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "beam_width": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA GB300" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "head_dim": { + "distinct": 5, + "max": 512, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "isl": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "vllm_flashinfer_flashinfertrtllmapidecode", + "vllm_triton_attn" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 10153, + "max": 1.9112159411112468, + "min": 0.008522666369875273, + "nulls": 0, + "type": "double" + }, + "num_heads": { + "distinct": 17, + "max": 96, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_key_value_heads": { + "distinct": 9, + "max": 96, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "generation_attention" + ] + }, + "step": { + "distinct": 17, + "max": 131071, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + }, + "window_size": { + "distinct": 6, + "max": 8192, + "min": 0, + "nulls": 0, + "type": "int64" + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb300/attention/vllm/0.25.0/generation_attention_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "batch_size", + "isl", + "num_heads", + "num_key_value_heads", + "head_dim", + "beam_width", + "attn_dtype", + "kv_cache_dtype", + "step", + "window_size" + ], + "rows": 65408, + "sha256": "6f3e1b7415f795be084cf888eac7f42d73274195e85778c7213a9d42f6c785eb", + "sidecar_sha256": "a5794ac332016a0982121fb23cc94fb05350f154d1a8770f11569b23c8b6ca24" + }, + "mhc_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA GB300" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "hc_mult": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "hidden_size": { + "distinct": 2, + "max": 7168, + "min": 4096, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "vllm.model_executor.kernels.mhc.tilelang.mhc_post_tilelang", + "vllm.model_executor.kernels.mhc.tilelang.mhc_pre_tilelang" + ] + }, + "latency": { + "distinct": 105, + "max": 21.7786, + "min": 0.0083, + "nulls": 0, + "type": "double" + }, + "num_sites": { + "distinct": 1, + "max": 2, + "min": 2, + "nulls": 0, + "type": "int64" + }, + "num_tokens": { + "distinct": 35, + "max": 524288, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "post", + "pre" + ] + }, + "sinkhorn_iters": { + "distinct": 1, + "max": 20, + "min": 20, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mhc/vllm/0.25.0/mhc_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "architecture", + "num_tokens", + "num_sites", + "hc_mult", + "hidden_size", + "sinkhorn_iters" + ], + "rows": 140, + "sha256": "82efd40dc53c876cab2f32d36a499e2688076111a8fd0faad7a55dec93d06f2c", + "sidecar_sha256": "aa54deb2dbdf3173fe328db70ec565b1eea966ba9d488c22bb825d556d3e85c2" + }, + "mla_bmm_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "bmm_dtype": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16" + ] + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA GB300" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "vllm_torch_bmm" + ] + }, + "latency": { + "distinct": 481, + "max": 0.5303296089172364, + "min": 0.004364800080657006, + "nulls": 0, + "type": "double" + }, + "num_heads": { + "distinct": 12, + "max": 128, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_tokens": { + "distinct": 28, + "max": 20480, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "mla_gen_post", + "mla_gen_pre" + ] + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mla_bmm/vllm/0.25.0/mla_bmm_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "bmm_dtype", + "num_tokens", + "num_heads" + ], + "rows": 636, + "sha256": "b98fcc689849e70362ed8cae68feda94d2afa0707fea8d3176ac1a6ef88a273d", + "sidecar_sha256": "865749cd01284bdee328aad42bf7bf70d958bc99b0cee8916191203686b195d1" + }, + "mla_context_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "DeepseekV3ForCausalLM" + ] + }, + "batch_size": { + "distinct": 9, + "max": 256, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA GB300" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 3, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8_block", + "nvfp4" + ] + }, + "isl": { + "distinct": 18, + "max": 32768, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 3, + "nulls": 0, + "type": "large_string", + "values": [ + "FLASHINFER_MLA", + "FLASH_ATTN", + "TRTLLM_RAGGED" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 8749, + "max": 110.3916, + "min": 0.0319, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "model": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "deepseek-ai/DeepSeek-V3" + ] + }, + "num_heads": { + "distinct": 12, + "max": 128, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "mla_context_module" + ] + }, + "step": { + "distinct": 1, + "max": 0, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mla/vllm/0.25.0/mla_context_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step" + ], + "rows": 13059, + "sha256": "71bbb588981ab59a86f16abe74a22da35862d7e2e6c22a750ff2c665961a64f5", + "sidecar_sha256": "df340d8b0c11b8a7b6abbe52054312e24990c9580b34dca78dca7615809c2234" + }, + "mla_generation_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV3ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA GB300" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 3, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16", + "fp8_block", + "nvfp4" + ] + }, + "isl": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHINFER_MLA" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 3440, + "max": 10.4203, + "min": 0.0303, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "deepseek-ai/DeepSeek-V3" + ] + }, + "num_heads": { + "distinct": 11, + "max": 128, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "mla_generation_module" + ] + }, + "step": { + "distinct": 17, + "max": 131072, + "min": 2, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mla/vllm/0.25.0/mla_generation_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step" + ], + "rows": 11400, + "sha256": "19fad0762bda00c3f6b7d729ced13bf5bc56961909e5c18b10859014709bb776", + "sidecar_sha256": "df340d8b0c11b8a7b6abbe52054312e24990c9580b34dca78dca7615809c2234" + }, + "moe_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA GB300" + ] + }, + "distribution": { + "distinct": 3, + "nulls": 0, + "type": "large_string", + "values": [ + "balanced", + "power_law_1.01", + "power_law_1.2" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "hidden_size": { + "distinct": 12, + "max": 8192, + "min": 1024, + "nulls": 0, + "type": "int64" + }, + "inter_size": { + "distinct": 16, + "max": 16384, + "min": 512, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 14, + "nulls": 0, + "type": "large_string", + "values": [ + "vllm_compressedtensorsw4a4mxfp4moe_marlin_marlinexperts_situ_as_silu", + "vllm_compressedtensorsw4a4nvfp4moe_flashinfer_cutlass_flashinferexperts", + "vllm_compressedtensorsw4a4nvfp4moe_flashinfer_trtllm_trtllmnvfp4expertsmodular", + "vllm_compressedtensorsw4a4nvfp4moe_flashinfer_trtllm_trtllmnvfp4expertsmonolithic", + "vllm_compressedtensorswna16marlinmoe_marlin_marlinexperts", + "vllm_fp8moe_flashinfer_trtllm_trtllmfp8expertsmodular", + "vllm_fp8moe_flashinfer_trtllm_trtllmfp8expertsmonolithic", + "vllm_gptossmxfp4moe_flashinfer_trtllm_mxfp4_bf16_trtllmmxfp4expertsmonolithic", + "vllm_modeloptfp8moe_flashinfer_cutlass_flashinferexperts", + "vllm_modeloptfp8moe_flashinfer_trtllm_trtllmfp8expertsmonolithic", + "vllm_mxfp4moe_flashinfer_trtllm_mxfp4_mxfp8_trtllmmxfp4expertsmodular", + "vllm_unquantizedfusedmoe_flashinfer_cutlass_flashinferexperts", + "vllm_unquantizedfusedmoe_flashinfer_trtllm_trtllmbf16experts", + "vllm_unquantizedfusedmoe_triton_tritonexperts" + ] + }, + "latency": { + "distinct": 55454, + "max": 63.608770751953124, + "min": 0.012563200294971466, + "nulls": 0, + "type": "double" + }, + "moe_dtype": { + "distinct": 7, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8", + "fp8_block", + "int4_wo", + "nvfp4", + "w4a16_mxfp4", + "w4a8_mxfp4_mxfp8" + ] + }, + "moe_ep_size": { + "distinct": 9, + "max": 256, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "moe_tp_size": { + "distinct": 6, + "max": 32, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_experts": { + "distinct": 10, + "max": 896, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "num_tokens": { + "distinct": 27, + "max": 16384, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "moe" + ] + }, + "topk": { + "distinct": 8, + "max": 22, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb300/moe/vllm/0.25.0/moe_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "moe_dtype", + "num_tokens", + "hidden_size", + "inter_size", + "topk", + "num_experts", + "moe_tp_size", + "moe_ep_size", + "distribution" + ], + "rows": 73002, + "sha256": "63c699e538e9b511a0af39c5419fbcc15b77724a35f380e383cff2966690f7d0", + "sidecar_sha256": "536c07582afe526e11cab40a9b56d69232cc79073d3ce342be4234bbf53980bb" + }, + "msa_context_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "MiniMaxM3ForCausalLM" + ] + }, + "batch_size": { + "distinct": 9, + "max": 256, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA GB300" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 3, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8_block", + "nvfp4" + ] + }, + "isl": { + "distinct": 18, + "max": 32768, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "MiniMaxM3SparseMSAImpl" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 9588, + "max": 101.8485, + "min": 0.0195, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "MiniMaxAI/MiniMax-M3" + ] + }, + "num_heads": { + "distinct": 12, + "max": 128, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "msa_context_module" + ] + }, + "step": { + "distinct": 2, + "max": 128, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/gb300/msa/vllm/0.25.0/msa_context_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step" + ], + "rows": 14174, + "sha256": "6609e6f089aec7394c09a727a319fd3194dca64ac81adf949ae1ce962e819743", + "sidecar_sha256": "8cb134772f538d3664355cfc6b3665aa8fe9c03daf0401b571a1359a0d7ed52a" + } + } + }, + "h100_sxm": { + "attempts": [ + { + "done": 37518, + "expected": 37518, + "failed": 0, + "id": "gemm-bfloat16", + "op": "gemm", + "plan_sha256": "37d5ec63f630c7a1fc241bcb3c59e2de5e52e68b54849496160dfedf0a12fe3a", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 37518, + "expected": 37518, + "failed": 0, + "id": "gemm-fp8", + "op": "gemm", + "plan_sha256": "a26b4b0dd1684df8b9645fddcb12b080ee89d9c39f0e08b3cb44f22926a5500e", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 9348, + "expected": 9633, + "failed": 285, + "id": "gemm-fp8_block-00", + "op": "gemm", + "plan_sha256": "e0f07c473cae107b8c2a5520c2bb1ae85045e499a268a73f0ba1ce97d035a42a", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 9348, + "expected": 9633, + "failed": 285, + "id": "gemm-fp8_block-01", + "op": "gemm", + "plan_sha256": "fee7715156b25ee24e67995e9ae17a3bede2bbeed890c479b99cc84ade4d8fc8", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 8856, + "expected": 9126, + "failed": 270, + "id": "gemm-fp8_block-02", + "op": "gemm", + "plan_sha256": "03318633ff86a012c7e9bf8310f97ba545972b973af130273420a0c381c85a66", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 8856, + "expected": 9126, + "failed": 270, + "id": "gemm-fp8_block-03", + "op": "gemm", + "plan_sha256": "c5da28e5b245195dbf0816485f8230a16d02cf3aa39e8fed9dcffb8da85e87c8", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 53771, + "expected": 54468, + "failed": 697, + "id": "attention_context", + "op": "attention_context", + "plan_sha256": "04acb1ef5df69580c3028de4dbe34f5e7697450ce783366632247395a3ed3507", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 5820, + "expected": 5856, + "failed": 36, + "id": "mla_context_module", + "op": "mla_context_module", + "plan_sha256": "07e0ec5401fec176bb9e2ffe8d5ec19c54a2355f25ead16ff7e1960a279ca0df", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 66708, + "expected": 67636, + "failed": 928, + "id": "attention_generation", + "op": "attention_generation", + "plan_sha256": "12df4833c99374841693e9bec6da73b3e7a8300acb5ca88e92c5c43d4bb24cdc", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 8760, + "expected": 8760, + "failed": 0, + "id": "mla_generation_module", + "op": "mla_generation_module", + "plan_sha256": "266b8e03bae0501992a44a80eb57c80b2e1dee9e4aba26b673a247ec13e947e8", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 7679, + "expected": 7679, + "failed": 0, + "id": "encoder_attention", + "op": "encoder_attention", + "plan_sha256": "8a4f0e45a56a4b73354b445a9b3bccde722835cb6c6c133991182ba092eb1296", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 300, + "expected": 300, + "failed": 0, + "id": "mla_bmm_gen_pre", + "op": "mla_bmm_gen_pre", + "plan_sha256": "3928a5f969ae055ba17244c0d53b504e8253870ffcbf60696731071ddb98e1e5", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 336, + "expected": 336, + "failed": 0, + "id": "mla_bmm_gen_post", + "op": "mla_bmm_gen_post", + "plan_sha256": "faeaa20a84179b59df5283aa9a238572a431f4f16f8521470519147073fe4b53", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 11270, + "expected": 11712, + "failed": 442, + "id": "msa_context_module", + "op": "msa_context_module", + "plan_sha256": "2fa04db807438446fe3c239cafe10fc48f75b102de2146fa128bbbba094d9a90", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 2016, + "expected": 2190, + "failed": 174, + "id": "moe", + "op": "moe", + "plan_sha256": "86632275c959e534d7bc4d5ee15cd3b7cf60a75638a57992d14529f70f5e4b61", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 75, + "expected": 84, + "failed": 9, + "id": "gdn", + "op": "gdn", + "plan_sha256": "ba375a4c3a8dbbe8b77cec5cfd0f9dafd245e18b99ace763cc738c9645962dda", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 139, + "expected": 140, + "failed": 1, + "id": "mhc_module", + "op": "mhc_module", + "plan_sha256": "4251dd697ea2e18fcaeed92b1a7cbdf452549a75d3a38284340d5da8502e03a1", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 5789, + "expected": 5840, + "failed": 51, + "id": "dsv4_csa_context_module", + "op": "dsv4_csa_context_module", + "plan_sha256": "38a6cd429db806b370f13a9b847ad836440166e29d1e68c8e0c56da13396d5db", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 1544, + "expected": 1544, + "failed": 0, + "id": "dsv4_csa_generation_module", + "op": "dsv4_csa_generation_module", + "plan_sha256": "21f9356813701f48d6451510032ec2746c9a64e3433c7f9954a0156c187dcae0", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 5793, + "expected": 5840, + "failed": 47, + "id": "dsv4_hca_context_module", + "op": "dsv4_hca_context_module", + "plan_sha256": "d36420598033f22968df6d517a051324bdac1f448ff41be9aacf12beec3dd893", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 1544, + "expected": 1544, + "failed": 0, + "id": "dsv4_hca_generation_module", + "op": "dsv4_hca_generation_module", + "plan_sha256": "f6f9fe724d2fa5ad60fb6d0a162f93cd7901bd26093925a5fef0cbae9f310eb4", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 3918, + "expected": 3918, + "failed": 0, + "id": "dsv4_paged_mqa_logits_module", + "op": "dsv4_paged_mqa_logits_module", + "plan_sha256": "0b35d1427f37f58a74ae69d6503f0eecb1ba5d60499325b10e1dad10a8199250", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 3918, + "expected": 3918, + "failed": 0, + "id": "dsv4_hca_attn_module", + "op": "dsv4_hca_attn_module", + "plan_sha256": "0e844f577823301a0af954d0d20baab4642bb1c0368652998bc28160dfab816d", + "plan_source": "archived_case_plan", + "unattempted": 0 + } + ], + "cases": { + "done": 290824, + "expected": 294319, + "failed": 3495, + "unattempted": 0 + }, + "rows": 353449, + "table_count": 17, + "tables": { + "context_attention_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "attn_dtype": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16" + ] + }, + "batch_size": { + "distinct": 9, + "max": 256, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "beam_width": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA H100 80GB HBM3" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "head_dim": { + "distinct": 5, + "max": 512, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "isl": { + "distinct": 17, + "max": 16384, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "vllm_flash_attn_fa3", + "vllm_flash_attn_fa4" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 35061, + "max": 205.19938151041666, + "min": 0.010405333091815313, + "nulls": 0, + "type": "double" + }, + "num_heads": { + "distinct": 17, + "max": 96, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_key_value_heads": { + "distinct": 12, + "max": 96, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "context_attention" + ] + }, + "step": { + "distinct": 1, + "max": 0, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + }, + "window_size": { + "distinct": 6, + "max": 8192, + "min": 0, + "nulls": 0, + "type": "int64" + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/attention/vllm/0.25.0/context_attention_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "batch_size", + "isl", + "num_heads", + "num_key_value_heads", + "head_dim", + "beam_width", + "attn_dtype", + "kv_cache_dtype", + "step", + "window_size" + ], + "rows": 53771, + "sha256": "7bca62aa7d140efcd54dadb60ff7582de130f7693265c26a53a817e92a601f9a", + "sidecar_sha256": "9c22cf1a4cb9d50fad792d7e319e933b10cf57feae22c0fb90a46127ce12fc46" + }, + "dsv4_csa_context_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA H100 80GB HBM3" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 21, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 5417, + "max": 357.5721, + "min": 0.0702, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 5, + "max": 128, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_csa_context_module" + ] + }, + "step": { + "distinct": 24, + "max": 65535, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 4, + "max": 8, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_context_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 5789, + "sha256": "fd4664e02b104be261b238db5dd571544dc35018ffa6e5fdda700f68e0503a79", + "sidecar_sha256": "19017dfd3c09a7df0dc5dd4d47e01ff0d9324998f3fe4f93b999effe36bfe8aa" + }, + "dsv4_csa_generation_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA H100 80GB HBM3" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 1031, + "max": 1.3585, + "min": 0.0695, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 5, + "max": 128, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_csa_generation_module" + ] + }, + "step": { + "distinct": 21, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 4, + "max": 8, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_generation_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 1544, + "sha256": "c9a6ff6f081dee79ff3053d32a8e85d550ee7aaab0c3d080bb2d33d7ca4ab839", + "sidecar_sha256": "19017dfd3c09a7df0dc5dd4d47e01ff0d9324998f3fe4f93b999effe36bfe8aa" + }, + "dsv4_hca_attn_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA H100 80GB HBM3" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 16, + "max": 8192, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 2358, + "max": 14.2497, + "min": 0.01, + "nulls": 0, + "type": "double" + }, + "local_num_heads": { + "distinct": 2, + "max": 128, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 2, + "max": 128, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_hca_attn_module" + ] + }, + "step": { + "distinct": 26, + "max": 1048575, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_attn_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "local_num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 3918, + "sha256": "86ca415600d3f4b468d683a689403ca298739cc65967c22c66fc2108ae6aa2fe", + "sidecar_sha256": "19017dfd3c09a7df0dc5dd4d47e01ff0d9324998f3fe4f93b999effe36bfe8aa" + }, + "dsv4_hca_context_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA H100 80GB HBM3" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 21, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 5153, + "max": 189.128, + "min": 0.0442, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 5, + "max": 128, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_hca_context_module" + ] + }, + "step": { + "distinct": 24, + "max": 65535, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 4, + "max": 8, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_context_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 5793, + "sha256": "9b0d69b9a9634ffd2d58b8e38d402db292ab003f5d4a59cab77a15c983f13d0d", + "sidecar_sha256": "19017dfd3c09a7df0dc5dd4d47e01ff0d9324998f3fe4f93b999effe36bfe8aa" + }, + "dsv4_hca_generation_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA H100 80GB HBM3" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 896, + "max": 1.1302, + "min": 0.0441, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 5, + "max": 128, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_hca_generation_module" + ] + }, + "step": { + "distinct": 21, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 4, + "max": 8, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_generation_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 1544, + "sha256": "ddeed06756d6e93d3afa9d372587ed19edae41f6051435816b27fc265e7689a8", + "sidecar_sha256": "19017dfd3c09a7df0dc5dd4d47e01ff0d9324998f3fe4f93b999effe36bfe8aa" + }, + "dsv4_paged_mqa_logits_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA H100 80GB HBM3" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 16, + "max": 8192, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "vllm.utils.deep_gemm.fp8_fp4_paged_mqa_logits" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 1148, + "max": 21.4616, + "min": 0.0044, + "nulls": 0, + "type": "double" + }, + "local_num_heads": { + "distinct": 2, + "max": 128, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_e4m3" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 2, + "max": 128, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_paged_mqa_logits_module" + ] + }, + "step": { + "distinct": 26, + "max": 1048575, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_paged_mqa_logits_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "local_num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 3918, + "sha256": "8eaca4d81931d542e879f8e1cd9728dda466ecd5d9c19449eef7be5876e90270", + "sidecar_sha256": "19017dfd3c09a7df0dc5dd4d47e01ff0d9324998f3fe4f93b999effe36bfe8aa" + }, + "encoder_attention_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "attn_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "batch_size": { + "distinct": 7, + "max": 64, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA H100 80GB HBM3" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "head_dim": { + "distinct": 3, + "max": 80, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "isl": { + "distinct": 39, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "vllm_vit_flash_attn_fa3" + ] + }, + "latency": { + "distinct": 6423, + "max": 138.16303507486978, + "min": 0.008000000069538752, + "nulls": 0, + "type": "double" + }, + "num_heads": { + "distinct": 12, + "max": 32, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "encoder_attention" + ] + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/encoder_attention/vllm/0.25.0/encoder_attention_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "batch_size", + "isl", + "num_heads", + "head_dim", + "attn_dtype" + ], + "rows": 7679, + "sha256": "eefab58f3269f3656b2bb0142fe1d7fa9e6d34aeca9d21a79733f7c86d733c7a", + "sidecar_sha256": "fec7a291346a39b4f2726cf57105ee86337678fc7c7e048613aa16268846eca5" + }, + "gdn_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "d_conv": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "d_model": { + "distinct": 7, + "max": 8192, + "min": 1024, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA H100 80GB HBM3" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "vLLM" + ] + }, + "head_k_dim": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "head_v_dim": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 4, + "nulls": 0, + "type": "string", + "values": [ + "causal_conv1d_fn", + "causal_conv1d_update", + "chunk_gated_delta_rule_flashinfer", + "fused_recurrent_gated_delta_rule_packed_decode" + ] + }, + "latency": { + "distinct": 9601, + "max": 32.52016357421875, + "min": 0.001780479997396469, + "nulls": 0, + "type": "double" + }, + "model_name": { + "distinct": 9, + "nulls": 0, + "type": "string", + "values": [ + "Qwen/Qwen3.5-0.8B", + "Qwen/Qwen3.5-122B-A10B", + "Qwen/Qwen3.5-27B", + "Qwen/Qwen3.5-2B", + "Qwen/Qwen3.5-35B-A3B", + "Qwen/Qwen3.5-397B-A17B", + "Qwen/Qwen3.5-4B", + "Qwen/Qwen3.5-9B", + "Qwen/Qwen3.8-2.4T-A95B" + ] + }, + "num_k_heads": { + "distinct": 5, + "max": 16, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_tokens": { + "distinct": 22, + "max": 2097152, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_v_heads": { + "distinct": 11, + "max": 128, + "min": 2, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "gdn" + ] + }, + "phase": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "context", + "generation" + ] + }, + "seq_len": { + "distinct": 16, + "max": 32768, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/linear_attention/vllm/0.25.0/gdn_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "phase", + "batch_size", + "seq_len", + "num_tokens", + "d_model", + "d_conv", + "num_k_heads", + "head_k_dim", + "num_v_heads", + "head_v_dim", + "model_name" + ], + "rows": 10284, + "sha256": "b534f9b318565bb01016f8e6ae96445766798692285bea8a744b273dcf0c87b3", + "sidecar_sha256": "938ca39a855f2e2846adae01f41f13a92e252388ba2df64ec97979bb9768b6d1" + }, + "gemm_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA H100 80GB HBM3" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_dtype": { + "distinct": 3, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16", + "fp8", + "fp8_block" + ] + }, + "k": { + "distinct": 23, + "max": 65536, + "min": 32, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 5, + "nulls": 0, + "type": "string", + "values": [ + "CutlassFP8ScaledMMLinearKernel", + "CutlassFp8BlockScaledMMKernel", + "DeepGemmFp8BlockScaledMMKernel", + "FlashInferFp8BlockScaledMMKernel", + "torch.nn.functional.linear" + ] + }, + "latency": { + "distinct": 69913, + "max": 330.73700629340277, + "min": 0.0025048889219760895, + "nulls": 0, + "type": "double" + }, + "m": { + "distinct": 74, + "max": 32768, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "n": { + "distinct": 30, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "gemm" + ] + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/gemm/vllm/0.25.0/gemm_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "gemm_dtype", + "m", + "n", + "k" + ], + "rows": 111444, + "sha256": "bdec76284417f792f5b05460ee6c4441cd5d71f775cc708fdfd7bb04299e2b3c", + "sidecar_sha256": "84860a00d7ecc3d8ff110d37e797f6945b8f06b260545d1d5671810dfa3e7bce" + }, + "generation_attention_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "attn_dtype": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16" + ] + }, + "batch_size": { + "distinct": 12, + "max": 2048, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "beam_width": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA H100 80GB HBM3" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "head_dim": { + "distinct": 5, + "max": 512, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "isl": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "vllm_flash_attn_fa3", + "vllm_flash_attn_fa4" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 19342, + "max": 9.665440241495768, + "min": 0.010384000216921171, + "nulls": 0, + "type": "double" + }, + "num_heads": { + "distinct": 17, + "max": 96, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_key_value_heads": { + "distinct": 9, + "max": 96, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "generation_attention" + ] + }, + "step": { + "distinct": 17, + "max": 131071, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + }, + "window_size": { + "distinct": 6, + "max": 8192, + "min": 0, + "nulls": 0, + "type": "int64" + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/attention/vllm/0.25.0/generation_attention_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "batch_size", + "isl", + "num_heads", + "num_key_value_heads", + "head_dim", + "beam_width", + "attn_dtype", + "kv_cache_dtype", + "step", + "window_size" + ], + "rows": 66708, + "sha256": "bb71db9055176c4b382bc6e0199df95fbbd56439345d4d9f4812f1bbf344a627", + "sidecar_sha256": "9c22cf1a4cb9d50fad792d7e319e933b10cf57feae22c0fb90a46127ce12fc46" + }, + "mhc_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA H100 80GB HBM3" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "hc_mult": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "hidden_size": { + "distinct": 2, + "max": 7168, + "min": 4096, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "vllm.model_executor.kernels.mhc.tilelang.mhc_post_tilelang", + "vllm.model_executor.kernels.mhc.tilelang.mhc_pre_tilelang" + ] + }, + "latency": { + "distinct": 128, + "max": 45.3827, + "min": 0.0064, + "nulls": 0, + "type": "double" + }, + "num_sites": { + "distinct": 1, + "max": 2, + "min": 2, + "nulls": 0, + "type": "int64" + }, + "num_tokens": { + "distinct": 35, + "max": 524288, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "post", + "pre" + ] + }, + "sinkhorn_iters": { + "distinct": 1, + "max": 20, + "min": 20, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mhc/vllm/0.25.0/mhc_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "architecture", + "num_tokens", + "num_sites", + "hc_mult", + "hidden_size", + "sinkhorn_iters" + ], + "rows": 139, + "sha256": "cb2772f76ee005632e8ff15011c94c9d59c62000484711d2e3f4189be6e8458f", + "sidecar_sha256": "b787c4924203d316bef7962c4aa3579aa50e8e985a3b69324a7340cae808e612" + }, + "mla_bmm_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "bmm_dtype": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16" + ] + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA H100 80GB HBM3" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "vllm_torch_bmm" + ] + }, + "latency": { + "distinct": 543, + "max": 1.156383991241455, + "min": 0.004239999875426293, + "nulls": 0, + "type": "double" + }, + "num_heads": { + "distinct": 12, + "max": 128, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_tokens": { + "distinct": 28, + "max": 20480, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "mla_gen_post", + "mla_gen_pre" + ] + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mla_bmm/vllm/0.25.0/mla_bmm_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "bmm_dtype", + "num_tokens", + "num_heads" + ], + "rows": 636, + "sha256": "646a21962b21e84601d629d9e11bbb03955000f90e28fc845daac8c7c4fdbcf1", + "sidecar_sha256": "6f5a0a5f75f1917a41749c48aca2d25cd0a7406b34db93d92359a305d555cd31" + }, + "mla_context_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "DeepseekV3ForCausalLM" + ] + }, + "batch_size": { + "distinct": 9, + "max": 256, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA H100 80GB HBM3" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8_block" + ] + }, + "isl": { + "distinct": 18, + "max": 32768, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 3, + "nulls": 0, + "type": "large_string", + "values": [ + "FLASHMLA", + "FLASH_ATTN", + "FLASH_ATTN_MLA" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 5004, + "max": 274.1819, + "min": 0.0367, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "deepseek-ai/DeepSeek-V3" + ] + }, + "num_heads": { + "distinct": 12, + "max": 128, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "mla_context_module" + ] + }, + "step": { + "distinct": 1, + "max": 0, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mla/vllm/0.25.0/mla_context_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step" + ], + "rows": 5820, + "sha256": "f6094e0ec5a3b46719c37cfcc4d0655de581e238116e22cdbc644613eca44728", + "sidecar_sha256": "42ae014b12646b8ed61a44d4be12127447dfc50d7025bb8196f18cd6494fe97e" + }, + "mla_generation_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV3ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA H100 80GB HBM3" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16", + "fp8_block" + ] + }, + "isl": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA", + "FLASH_ATTN_MLA" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 3716, + "max": 12.9695, + "min": 0.0358, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "deepseek-ai/DeepSeek-V3" + ] + }, + "num_heads": { + "distinct": 12, + "max": 128, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "mla_generation_module" + ] + }, + "step": { + "distinct": 17, + "max": 131072, + "min": 2, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mla/vllm/0.25.0/mla_generation_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step" + ], + "rows": 8760, + "sha256": "37066dfe8c1f39411c9d14582facbf00c5d063db623848982cf0d7abd443ade2", + "sidecar_sha256": "42ae014b12646b8ed61a44d4be12127447dfc50d7025bb8196f18cd6494fe97e" + }, + "moe_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA H100 80GB HBM3" + ] + }, + "distribution": { + "distinct": 3, + "nulls": 0, + "type": "large_string", + "values": [ + "balanced", + "power_law_1.01", + "power_law_1.2" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "hidden_size": { + "distinct": 12, + "max": 8192, + "min": 1024, + "nulls": 0, + "type": "int64" + }, + "inter_size": { + "distinct": 16, + "max": 16384, + "min": 512, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 7, + "nulls": 0, + "type": "large_string", + "values": [ + "vllm_compressedtensorsw4a4mxfp4moe_marlin_marlinexperts_situ_as_silu", + "vllm_compressedtensorswna16marlinmoe_marlin_marlinexperts", + "vllm_fp8moe_flashinfer_cutlass_flashinferexperts", + "vllm_fp8moe_triton_tritonexperts", + "vllm_gptossmxfp4moe_triton_oaitritonmxfp4expertsmonolithic", + "vllm_modeloptfp8moe_flashinfer_cutlass_flashinferexperts", + "vllm_unquantizedfusedmoe_triton_tritonexperts" + ] + }, + "latency": { + "distinct": 50748, + "max": 104.4814453125, + "min": 0.02351360023021698, + "nulls": 0, + "type": "double" + }, + "moe_dtype": { + "distinct": 5, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8", + "fp8_block", + "int4_wo", + "w4a16_mxfp4" + ] + }, + "moe_ep_size": { + "distinct": 9, + "max": 256, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "moe_tp_size": { + "distinct": 6, + "max": 32, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_experts": { + "distinct": 10, + "max": 896, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "num_tokens": { + "distinct": 27, + "max": 16384, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "moe" + ] + }, + "topk": { + "distinct": 8, + "max": 22, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/moe/vllm/0.25.0/moe_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "moe_dtype", + "num_tokens", + "hidden_size", + "inter_size", + "topk", + "num_experts", + "moe_tp_size", + "moe_ep_size", + "distribution" + ], + "rows": 54432, + "sha256": "33d30aa589ab9d525669888c033a524ce348f53f99798c0a22e29c49a462911b", + "sidecar_sha256": "cb82bcf27db8b4dd1c4516f9a5054bedb65729e2c35a662c7166499bd26f5e1a" + }, + "msa_context_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "MiniMaxM3ForCausalLM" + ] + }, + "batch_size": { + "distinct": 9, + "max": 256, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA H100 80GB HBM3" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8_block" + ] + }, + "isl": { + "distinct": 18, + "max": 32768, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "MiniMaxM3SparseTritonImpl" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 8980, + "max": 326.3513, + "min": 0.0221, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "MiniMaxAI/MiniMax-M3" + ] + }, + "num_heads": { + "distinct": 12, + "max": 128, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "msa_context_module" + ] + }, + "step": { + "distinct": 2, + "max": 128, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/msa/vllm/0.25.0/msa_context_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step" + ], + "rows": 11270, + "sha256": "5ae6688e485d4e160569a602abfad2abd9848edf198abcfe7f271be88057dea7", + "sidecar_sha256": "7cd4884fcfca85da911b57175ada3500b2d1f6f45eaae416f68c956c5e5ee790" + } + } + }, + "h200_sxm": { + "attempts": [ + { + "done": 37518, + "expected": 37518, + "failed": 0, + "id": "gemm-bfloat16", + "op": "gemm", + "plan_sha256": "37d5ec63f630c7a1fc241bcb3c59e2de5e52e68b54849496160dfedf0a12fe3a", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 37518, + "expected": 37518, + "failed": 0, + "id": "gemm-fp8", + "op": "gemm", + "plan_sha256": "a26b4b0dd1684df8b9645fddcb12b080ee89d9c39f0e08b3cb44f22926a5500e", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 9348, + "expected": 9633, + "failed": 285, + "id": "gemm-fp8_block-00", + "op": "gemm", + "plan_sha256": "e0f07c473cae107b8c2a5520c2bb1ae85045e499a268a73f0ba1ce97d035a42a", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 9348, + "expected": 9633, + "failed": 285, + "id": "gemm-fp8_block-01", + "op": "gemm", + "plan_sha256": "fee7715156b25ee24e67995e9ae17a3bede2bbeed890c479b99cc84ade4d8fc8", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 8856, + "expected": 9126, + "failed": 270, + "id": "gemm-fp8_block-02", + "op": "gemm", + "plan_sha256": "03318633ff86a012c7e9bf8310f97ba545972b973af130273420a0c381c85a66", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 8856, + "expected": 9126, + "failed": 270, + "id": "gemm-fp8_block-03", + "op": "gemm", + "plan_sha256": "c5da28e5b245195dbf0816485f8230a16d02cf3aa39e8fed9dcffb8da85e87c8", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 53771, + "expected": 54468, + "failed": 697, + "id": "attention_context", + "op": "attention_context", + "plan_sha256": "04acb1ef5df69580c3028de4dbe34f5e7697450ce783366632247395a3ed3507", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 5820, + "expected": 5856, + "failed": 36, + "id": "mla_context_module", + "op": "mla_context_module", + "plan_sha256": "07e0ec5401fec176bb9e2ffe8d5ec19c54a2355f25ead16ff7e1960a279ca0df", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 66708, + "expected": 67636, + "failed": 928, + "id": "attention_generation", + "op": "attention_generation", + "plan_sha256": "12df4833c99374841693e9bec6da73b3e7a8300acb5ca88e92c5c43d4bb24cdc", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 8832, + "expected": 8832, + "failed": 0, + "id": "mla_generation_module", + "op": "mla_generation_module", + "plan_sha256": "23f8a42b14c9eec372c32c809e900acb3fb4e5e1bad5c9adf14124dacd32943f", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 7679, + "expected": 7679, + "failed": 0, + "id": "encoder_attention", + "op": "encoder_attention", + "plan_sha256": "8a4f0e45a56a4b73354b445a9b3bccde722835cb6c6c133991182ba092eb1296", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 300, + "expected": 300, + "failed": 0, + "id": "mla_bmm_gen_pre", + "op": "mla_bmm_gen_pre", + "plan_sha256": "3928a5f969ae055ba17244c0d53b504e8253870ffcbf60696731071ddb98e1e5", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 336, + "expected": 336, + "failed": 0, + "id": "mla_bmm_gen_post", + "op": "mla_bmm_gen_post", + "plan_sha256": "faeaa20a84179b59df5283aa9a238572a431f4f16f8521470519147073fe4b53", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 11268, + "expected": 11712, + "failed": 444, + "id": "msa_context_module", + "op": "msa_context_module", + "plan_sha256": "2fa04db807438446fe3c239cafe10fc48f75b102de2146fa128bbbba094d9a90", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 2016, + "expected": 2190, + "failed": 174, + "id": "moe", + "op": "moe", + "plan_sha256": "86632275c959e534d7bc4d5ee15cd3b7cf60a75638a57992d14529f70f5e4b61", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 79, + "expected": 84, + "failed": 5, + "id": "gdn", + "op": "gdn", + "plan_sha256": "ba375a4c3a8dbbe8b77cec5cfd0f9dafd245e18b99ace763cc738c9645962dda", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 140, + "expected": 140, + "failed": 0, + "id": "mhc_module", + "op": "mhc_module", + "plan_sha256": "4251dd697ea2e18fcaeed92b1a7cbdf452549a75d3a38284340d5da8502e03a1", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 5840, + "expected": 5840, + "failed": 0, + "id": "dsv4_csa_context_module", + "op": "dsv4_csa_context_module", + "plan_sha256": "38a6cd429db806b370f13a9b847ad836440166e29d1e68c8e0c56da13396d5db", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 1544, + "expected": 1544, + "failed": 0, + "id": "dsv4_csa_generation_module", + "op": "dsv4_csa_generation_module", + "plan_sha256": "21f9356813701f48d6451510032ec2746c9a64e3433c7f9954a0156c187dcae0", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 5840, + "expected": 5840, + "failed": 0, + "id": "dsv4_hca_context_module", + "op": "dsv4_hca_context_module", + "plan_sha256": "d36420598033f22968df6d517a051324bdac1f448ff41be9aacf12beec3dd893", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 1544, + "expected": 1544, + "failed": 0, + "id": "dsv4_hca_generation_module", + "op": "dsv4_hca_generation_module", + "plan_sha256": "f6f9fe724d2fa5ad60fb6d0a162f93cd7901bd26093925a5fef0cbae9f310eb4", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 3918, + "expected": 3918, + "failed": 0, + "id": "dsv4_paged_mqa_logits_module", + "op": "dsv4_paged_mqa_logits_module", + "plan_sha256": "0b35d1427f37f58a74ae69d6503f0eecb1ba5d60499325b10e1dad10a8199250", + "plan_source": "archived_case_plan", + "unattempted": 0 + }, + { + "done": 3918, + "expected": 3918, + "failed": 0, + "id": "dsv4_hca_attn_module", + "op": "dsv4_hca_attn_module", + "plan_sha256": "0e844f577823301a0af954d0d20baab4642bb1c0368652998bc28160dfab816d", + "plan_source": "archived_case_plan", + "unattempted": 0 + } + ], + "cases": { + "done": 290997, + "expected": 294391, + "failed": 3394, + "unattempted": 0 + }, + "rows": 353658, + "table_count": 17, + "tables": { + "context_attention_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "attn_dtype": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16" + ] + }, + "batch_size": { + "distinct": 9, + "max": 256, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "beam_width": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA H200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "head_dim": { + "distinct": 5, + "max": 512, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "isl": { + "distinct": 17, + "max": 16384, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "vllm_flash_attn_fa3", + "vllm_flash_attn_fa4" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 34855, + "max": 204.6250203450521, + "min": 0.00966933307548364, + "nulls": 0, + "type": "double" + }, + "num_heads": { + "distinct": 17, + "max": 96, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_key_value_heads": { + "distinct": 12, + "max": 96, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "context_attention" + ] + }, + "step": { + "distinct": 1, + "max": 0, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + }, + "window_size": { + "distinct": 6, + "max": 8192, + "min": 0, + "nulls": 0, + "type": "int64" + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/attention/vllm/0.25.0/context_attention_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "batch_size", + "isl", + "num_heads", + "num_key_value_heads", + "head_dim", + "beam_width", + "attn_dtype", + "kv_cache_dtype", + "step", + "window_size" + ], + "rows": 53771, + "sha256": "b5935611d339f2ca02c4c9b671d403842ecc26a9571f8c4f3ad70f378aabeb03", + "sidecar_sha256": "9c22cf1a4cb9d50fad792d7e319e933b10cf57feae22c0fb90a46127ce12fc46" + }, + "dsv4_csa_context_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA H200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 21, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 5445, + "max": 493.2763, + "min": 0.0634, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 5, + "max": 128, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_csa_context_module" + ] + }, + "step": { + "distinct": 24, + "max": 65535, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 4, + "max": 8, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_context_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 5840, + "sha256": "8fec19f573b830fd854b0feb199cc6841f8070a86308dc892eec80afdaf93c29", + "sidecar_sha256": "23f9cb39768514c0ef6f079e9bc43abc2719cd464638e371fda85f2922ce7bc1" + }, + "dsv4_csa_generation_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA H200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 938, + "max": 1.288, + "min": 0.0641, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 5, + "max": 128, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_csa_generation_module" + ] + }, + "step": { + "distinct": 21, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 4, + "max": 8, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_generation_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 1544, + "sha256": "a4c1dc94f505ba0360125de51735948185b0e9bd44128a42668613875ae7741a", + "sidecar_sha256": "23f9cb39768514c0ef6f079e9bc43abc2719cd464638e371fda85f2922ce7bc1" + }, + "dsv4_hca_attn_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA H200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 16, + "max": 8192, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 2307, + "max": 14.4426, + "min": 0.0096, + "nulls": 0, + "type": "double" + }, + "local_num_heads": { + "distinct": 2, + "max": 128, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 2, + "max": 128, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_hca_attn_module" + ] + }, + "step": { + "distinct": 26, + "max": 1048575, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_attn_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "local_num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 3918, + "sha256": "3b0d87916ae3f6414cbe92966632b9133fb162a596ffce6b44cfeb916da3a4af", + "sidecar_sha256": "23f9cb39768514c0ef6f079e9bc43abc2719cd464638e371fda85f2922ce7bc1" + }, + "dsv4_hca_context_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA H200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 21, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 5143, + "max": 285.9865, + "min": 0.042, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 5, + "max": 128, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_hca_context_module" + ] + }, + "step": { + "distinct": 24, + "max": 65535, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 4, + "max": 8, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_context_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 5840, + "sha256": "39fa711b357b5c72251862cc741ae81e6a486f6ad0adb2ae29e25565d99642b1", + "sidecar_sha256": "23f9cb39768514c0ef6f079e9bc43abc2719cd464638e371fda85f2922ce7bc1" + }, + "dsv4_hca_generation_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA H200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA_SPARSE_DSV4" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 844, + "max": 1.0666, + "min": 0.0416, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 5, + "max": 128, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_hca_generation_module" + ] + }, + "step": { + "distinct": 21, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 4, + "max": 8, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_generation_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 1544, + "sha256": "d9683b5d8a43df2f023bfb4ac50d38f8717b0c447d2ed3962b3ae5eaa64fbf00", + "sidecar_sha256": "23f9cb39768514c0ef6f079e9bc43abc2719cd464638e371fda85f2922ce7bc1" + }, + "dsv4_paged_mqa_logits_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "compress_ratio": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA H200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_block" + ] + }, + "isl": { + "distinct": 16, + "max": 8192, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "vllm.utils.deep_gemm.fp8_fp4_paged_mqa_logits" + ] + }, + "kv_cache_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8" + ] + }, + "latency": { + "distinct": 1088, + "max": 22.3937, + "min": 0.004, + "nulls": 0, + "type": "double" + }, + "local_num_heads": { + "distinct": 2, + "max": 128, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "fp8_e4m3" + ] + }, + "model": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "sgl-project/DeepSeek-V4-Flash-FP8", + "sgl-project/DeepSeek-V4-Pro-FP8" + ] + }, + "num_heads": { + "distinct": 2, + "max": 128, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "dsv4_paged_mqa_logits_module" + ] + }, + "step": { + "distinct": 26, + "max": 1048575, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_paged_mqa_logits_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "local_num_heads", + "batch_size", + "isl", + "tp_size", + "step", + "compress_ratio" + ], + "rows": 3918, + "sha256": "ad63707a693a23b190807366e1c6eb65cacc33ef42a92b070c23f6985ed0514b", + "sidecar_sha256": "23f9cb39768514c0ef6f079e9bc43abc2719cd464638e371fda85f2922ce7bc1" + }, + "encoder_attention_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "attn_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "batch_size": { + "distinct": 7, + "max": 64, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA H200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "head_dim": { + "distinct": 3, + "max": 80, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "isl": { + "distinct": 39, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "vllm_vit_flash_attn_fa3" + ] + }, + "latency": { + "distinct": 6397, + "max": 134.47529093424478, + "min": 0.007354666789372762, + "nulls": 0, + "type": "double" + }, + "num_heads": { + "distinct": 12, + "max": 32, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "encoder_attention" + ] + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/encoder_attention/vllm/0.25.0/encoder_attention_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "batch_size", + "isl", + "num_heads", + "head_dim", + "attn_dtype" + ], + "rows": 7679, + "sha256": "a119a81e8d6b73d71d429eeabd4aaa5e942e77acb267f5f74b4e580c18d028d2", + "sidecar_sha256": "fec7a291346a39b4f2726cf57105ee86337678fc7c7e048613aa16268846eca5" + }, + "gdn_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "d_conv": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "d_model": { + "distinct": 7, + "max": 8192, + "min": 1024, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA H200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "vLLM" + ] + }, + "head_k_dim": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "head_v_dim": { + "distinct": 1, + "max": 128, + "min": 128, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 4, + "nulls": 0, + "type": "string", + "values": [ + "causal_conv1d_fn", + "causal_conv1d_update", + "chunk_gated_delta_rule_flashinfer", + "fused_recurrent_gated_delta_rule_packed_decode" + ] + }, + "latency": { + "distinct": 9614, + "max": 57.375117187499995, + "min": 0.0016345599293708802, + "nulls": 0, + "type": "double" + }, + "model_name": { + "distinct": 9, + "nulls": 0, + "type": "string", + "values": [ + "Qwen/Qwen3.5-0.8B", + "Qwen/Qwen3.5-122B-A10B", + "Qwen/Qwen3.5-27B", + "Qwen/Qwen3.5-2B", + "Qwen/Qwen3.5-35B-A3B", + "Qwen/Qwen3.5-397B-A17B", + "Qwen/Qwen3.5-4B", + "Qwen/Qwen3.5-9B", + "Qwen/Qwen3.8-2.4T-A95B" + ] + }, + "num_k_heads": { + "distinct": 5, + "max": 16, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_tokens": { + "distinct": 22, + "max": 2097152, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_v_heads": { + "distinct": 11, + "max": 128, + "min": 2, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "gdn" + ] + }, + "phase": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "context", + "generation" + ] + }, + "seq_len": { + "distinct": 16, + "max": 32768, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/linear_attention/vllm/0.25.0/gdn_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "phase", + "batch_size", + "seq_len", + "num_tokens", + "d_model", + "d_conv", + "num_k_heads", + "head_k_dim", + "num_v_heads", + "head_v_dim", + "model_name" + ], + "rows": 10324, + "sha256": "1b3108961ea6f21f2901c1983a8d6ef80a91f5a3793408044c5c791a0e4187b9", + "sidecar_sha256": "18aed8622a28b426c35cd62e605e3f1f092f3f7e494594b8cfb3cefc9139512d" + }, + "gemm_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA H200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_dtype": { + "distinct": 3, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16", + "fp8", + "fp8_block" + ] + }, + "k": { + "distinct": 23, + "max": 65536, + "min": 32, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 5, + "nulls": 0, + "type": "string", + "values": [ + "CutlassFP8ScaledMMLinearKernel", + "CutlassFp8BlockScaledMMKernel", + "DeepGemmFp8BlockScaledMMKernel", + "FlashInferFp8BlockScaledMMKernel", + "torch.nn.functional.linear" + ] + }, + "latency": { + "distinct": 68012, + "max": 268.51917860243054, + "min": 0.0024293334119849736, + "nulls": 0, + "type": "double" + }, + "m": { + "distinct": 74, + "max": 32768, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "n": { + "distinct": 30, + "max": 65536, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "gemm" + ] + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/gemm/vllm/0.25.0/gemm_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "gemm_dtype", + "m", + "n", + "k" + ], + "rows": 111444, + "sha256": "ab1046cb36d91f3adb01e0806763489626f4ec53a438b0e932a45256d3875243", + "sidecar_sha256": "8e0f533ea11eb2cb1ec144a059fd4d51f82f73d04cad9f364929f74d64a5ff42" + }, + "generation_attention_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "attn_dtype": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16" + ] + }, + "batch_size": { + "distinct": 12, + "max": 2048, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "beam_width": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA H200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "head_dim": { + "distinct": 5, + "max": 512, + "min": 64, + "nulls": 0, + "type": "int64" + }, + "isl": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "vllm_flash_attn_fa3", + "vllm_flash_attn_fa4" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 18847, + "max": 9.667919794718424, + "min": 0.00966933307548364, + "nulls": 0, + "type": "double" + }, + "num_heads": { + "distinct": 17, + "max": 96, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_key_value_heads": { + "distinct": 9, + "max": 96, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "generation_attention" + ] + }, + "step": { + "distinct": 17, + "max": 131071, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + }, + "window_size": { + "distinct": 6, + "max": 8192, + "min": 0, + "nulls": 0, + "type": "int64" + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/attention/vllm/0.25.0/generation_attention_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "batch_size", + "isl", + "num_heads", + "num_key_value_heads", + "head_dim", + "beam_width", + "attn_dtype", + "kv_cache_dtype", + "step", + "window_size" + ], + "rows": 66708, + "sha256": "356b18abc323349058377f1aa4652c672b048b6e24d0b54bcd12fdd9cdb5284c", + "sidecar_sha256": "9c22cf1a4cb9d50fad792d7e319e933b10cf57feae22c0fb90a46127ce12fc46" + }, + "mhc_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV4ForCausalLM" + ] + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA H200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "hc_mult": { + "distinct": 1, + "max": 4, + "min": 4, + "nulls": 0, + "type": "int64" + }, + "hidden_size": { + "distinct": 2, + "max": 7168, + "min": 4096, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "vllm.model_executor.kernels.mhc.tilelang.mhc_post_tilelang", + "vllm.model_executor.kernels.mhc.tilelang.mhc_pre_tilelang" + ] + }, + "latency": { + "distinct": 125, + "max": 33.1729, + "min": 0.0061, + "nulls": 0, + "type": "double" + }, + "num_sites": { + "distinct": 1, + "max": 2, + "min": 2, + "nulls": 0, + "type": "int64" + }, + "num_tokens": { + "distinct": 35, + "max": 524288, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "post", + "pre" + ] + }, + "sinkhorn_iters": { + "distinct": 1, + "max": 20, + "min": 20, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mhc/vllm/0.25.0/mhc_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "architecture", + "num_tokens", + "num_sites", + "hc_mult", + "hidden_size", + "sinkhorn_iters" + ], + "rows": 140, + "sha256": "739f46c260f2f6d75a65d48310d482c47405b276a4b11b32a4a81c3ce3bfe7d3", + "sidecar_sha256": "ffe0f95d513b11f868ecc42c66e696fd16513f3afc98bde9982278c4cf444376" + }, + "mla_bmm_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "bmm_dtype": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16" + ] + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA H200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "vllm_torch_bmm" + ] + }, + "latency": { + "distinct": 551, + "max": 0.9571392059326171, + "min": 0.00352960005402565, + "nulls": 0, + "type": "double" + }, + "num_heads": { + "distinct": 12, + "max": 128, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_tokens": { + "distinct": 28, + "max": 20480, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "mla_gen_post", + "mla_gen_pre" + ] + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mla_bmm/vllm/0.25.0/mla_bmm_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "bmm_dtype", + "num_tokens", + "num_heads" + ], + "rows": 636, + "sha256": "9216eec0777bcedcc0fc5eb261d01a2f90a410eed92a0f8cc1fdc76edc3c3020", + "sidecar_sha256": "6f5a0a5f75f1917a41749c48aca2d25cd0a7406b34db93d92359a305d555cd31" + }, + "mla_context_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "DeepseekV3ForCausalLM" + ] + }, + "batch_size": { + "distinct": 9, + "max": 256, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA H200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8_block" + ] + }, + "isl": { + "distinct": 18, + "max": 32768, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 3, + "nulls": 0, + "type": "large_string", + "values": [ + "FLASHMLA", + "FLASH_ATTN", + "FLASH_ATTN_MLA" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 4969, + "max": 284.0853, + "min": 0.034, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "deepseek-ai/DeepSeek-V3" + ] + }, + "num_heads": { + "distinct": 12, + "max": 128, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "mla_context_module" + ] + }, + "step": { + "distinct": 1, + "max": 0, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mla/vllm/0.25.0/mla_context_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step" + ], + "rows": 5820, + "sha256": "dc9f9b7546b383d60ea454a0c2057ed84be8240d6241ffdb153e4cc9fc9afc61", + "sidecar_sha256": "661cd031f809a1e22e52ced5187e24b8d6534cc1edd7162b452d54febdffe241" + }, + "mla_generation_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "DeepseekV3ForCausalLM" + ] + }, + "batch_size": { + "distinct": 11, + "max": 1024, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "NVIDIA H200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16", + "fp8_block" + ] + }, + "isl": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "FLASHMLA", + "FLASH_ATTN_MLA" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 3556, + "max": 13.3992, + "min": 0.0339, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "deepseek-ai/DeepSeek-V3" + ] + }, + "num_heads": { + "distinct": 12, + "max": 128, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "mla_generation_module" + ] + }, + "step": { + "distinct": 17, + "max": 131072, + "min": 2, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mla/vllm/0.25.0/mla_generation_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step" + ], + "rows": 8832, + "sha256": "f8bc1fd52a3c3aed7839a9694c889c563108a30db43649f279c06c2d69f2c5bf", + "sidecar_sha256": "661cd031f809a1e22e52ced5187e24b8d6534cc1edd7162b452d54febdffe241" + }, + "moe_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA H200" + ] + }, + "distribution": { + "distinct": 3, + "nulls": 0, + "type": "large_string", + "values": [ + "balanced", + "power_law_1.01", + "power_law_1.2" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "hidden_size": { + "distinct": 12, + "max": 8192, + "min": 1024, + "nulls": 0, + "type": "int64" + }, + "inter_size": { + "distinct": 16, + "max": 16384, + "min": 512, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 7, + "nulls": 0, + "type": "large_string", + "values": [ + "vllm_compressedtensorsw4a4mxfp4moe_marlin_marlinexperts_situ_as_silu", + "vllm_compressedtensorswna16marlinmoe_marlin_marlinexperts", + "vllm_fp8moe_flashinfer_cutlass_flashinferexperts", + "vllm_fp8moe_triton_tritonexperts", + "vllm_gptossmxfp4moe_triton_oaitritonmxfp4expertsmonolithic", + "vllm_modeloptfp8moe_flashinfer_cutlass_flashinferexperts", + "vllm_unquantizedfusedmoe_triton_tritonexperts" + ] + }, + "latency": { + "distinct": 50129, + "max": 79.26846923828126, + "min": 0.02152319997549057, + "nulls": 0, + "type": "double" + }, + "moe_dtype": { + "distinct": 5, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8", + "fp8_block", + "int4_wo", + "w4a16_mxfp4" + ] + }, + "moe_ep_size": { + "distinct": 9, + "max": 256, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "moe_tp_size": { + "distinct": 6, + "max": 32, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "num_experts": { + "distinct": 10, + "max": 896, + "min": 8, + "nulls": 0, + "type": "int64" + }, + "num_tokens": { + "distinct": 27, + "max": 16384, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "moe" + ] + }, + "topk": { + "distinct": 8, + "max": 22, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/moe/vllm/0.25.0/moe_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "moe_dtype", + "num_tokens", + "hidden_size", + "inter_size", + "topk", + "num_experts", + "moe_tp_size", + "moe_ep_size", + "distribution" + ], + "rows": 54432, + "sha256": "eec0eba2d64eef570b1daa404eb5803f31a01156491791c88ba65e1696872732", + "sidecar_sha256": "e87476677af6e4da33c73ac1ae0a48f77551fbb002c9ed57cbf21244560f6a04" + }, + "msa_context_module_perf.parquet": { + "anomalies": { + "duplicate_physical_keys": 0, + "nonfinite_latency": 0, + "nonpositive_latency": 0, + "null_cells": 0 + }, + "columns": { + "architecture": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "MiniMaxM3ForCausalLM" + ] + }, + "batch_size": { + "distinct": 9, + "max": 256, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "device": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "NVIDIA H200" + ] + }, + "framework": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "VLLM" + ] + }, + "gemm_type": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8_block" + ] + }, + "isl": { + "distinct": 18, + "max": 32768, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "kernel_source": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "MiniMaxM3SparseTritonImpl" + ] + }, + "kv_cache_dtype": { + "distinct": 2, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16", + "fp8" + ] + }, + "latency": { + "distinct": 8948, + "max": 332.7369, + "min": 0.0206, + "nulls": 0, + "type": "double" + }, + "mla_dtype": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "bfloat16" + ] + }, + "model": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "MiniMaxAI/MiniMax-M3" + ] + }, + "num_heads": { + "distinct": 12, + "max": 128, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "op_name": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "msa_context_module" + ] + }, + "step": { + "distinct": 2, + "max": 128, + "min": 0, + "nulls": 0, + "type": "int64" + }, + "tp_size": { + "distinct": 1, + "max": 1, + "min": 1, + "nulls": 0, + "type": "int64" + }, + "version": { + "distinct": 1, + "nulls": 0, + "type": "large_string", + "values": [ + "0.25.0" + ] + } + }, + "latency_unit": "ms", + "path": "python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/msa/vllm/0.25.0/msa_context_module_perf.parquet", + "physical_key_columns": [ + "framework", + "version", + "op_name", + "kernel_source", + "model", + "architecture", + "mla_dtype", + "kv_cache_dtype", + "gemm_type", + "num_heads", + "batch_size", + "isl", + "tp_size", + "step" + ], + "rows": 11268, + "sha256": "91a5e4edb129b8914ba2fbda78c6cedf6d256d61734a1e6947da42a57f5c385c", + "sidecar_sha256": "eb99a300e716c6ae4a23994cce5babb569d4e64607a61b7205be2a52828dee5e" + } + } + } + }, + "vllm_runtime_revision": "dd10e03f95f94edbea1975c67ace3a35ec9a8a40" +} diff --git a/python/aisimulate/docs/perf_database/collector-v3-op-centric-design.md b/python/aisimulate/docs/perf_database/collector-v3-op-centric-design.md index eb449d447e..7e3f5c495f 100644 --- a/python/aisimulate/docs/perf_database/collector-v3-op-centric-design.md +++ b/python/aisimulate/docs/perf_database/collector-v3-op-centric-design.md @@ -322,6 +322,23 @@ have been invalidated by it, so backward fill preserves historical reproducibility; forward fill silently answers "how fast is 0.5.14" with 0.5.15's kernels. +### Corrected vLLM 0.24.0 FP8-block measurements + +For B200/B300/GB200/GB300 and H100/H200 SXM, the 0.24.0 GEMM tables +exclude FP8-block rows measured eagerly with host launch gaps (PR #219). +All other rows are retained unchanged. Their standard collection sidecars mark +the reduced tables partial, and existing declared reuse points to 0.25.0. +With shared-layer reuse enabled, explicit 0.24.0 requests load retained primary +rows first and fill missing keys from 0.25.0 graph-timed measurements. +Newer versions are never selected implicitly. With reuse disabled, the removed +FP8-block measurements remain unavailable. +The six-GPU regression test compares the entire native loaded table with the +first-source-wins merge of all primary and donor rows, including exact latency +preservation. Table-wide reuse also fills missing BF16, FP8, and (on Blackwell) +NVFP4 keys. The [PR #244 evidence bundle](../../../../docs/data/pr244/README.md) +records the per-precision counts, unchanged kernel source paths, cross-version +limitations, and reproducible collection coverage for all 85 new tables. + ### 6.3 Channel 2 — declared reuse (`reuse.yaml`, same backend, any direction) When we *know* data is valid for a version we never collected — typically a diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b200_sxm/gemm/vllm/0.24.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/b200_sxm/gemm/vllm/0.24.0/collection_meta.yaml index da9d3df044..95c1897814 100644 --- a/python/aisimulate/src/aiconfigurator_core/systems/data/b200_sxm/gemm/vllm/0.24.0/collection_meta.yaml +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/b200_sxm/gemm/vllm/0.24.0/collection_meta.yaml @@ -8,4 +8,6 @@ runtime: version: 0.24.0 tables: gemm_perf: - status: complete + status: partial + rows: 107226 + # Eager FP8-block timings removed; reuse.yaml declares the 0.25.0 donor. diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b200_sxm/gemm/vllm/0.24.0/gemm_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/b200_sxm/gemm/vllm/0.24.0/gemm_perf.parquet index d3565f4458..3d701bd44f 100644 Binary files a/python/aisimulate/src/aiconfigurator_core/systems/data/b200_sxm/gemm/vllm/0.24.0/gemm_perf.parquet and b/python/aisimulate/src/aiconfigurator_core/systems/data/b200_sxm/gemm/vllm/0.24.0/gemm_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b200_sxm/gemm/vllm/0.24.0/reuse.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/b200_sxm/gemm/vllm/0.24.0/reuse.yaml new file mode 100644 index 0000000000..6d5644a69f --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/b200_sxm/gemm/vllm/0.24.0/reuse.yaml @@ -0,0 +1,9 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 1 +reuse: + - table: gemm_perf + from_version: "0.25.0" + reason: "Fill missing GEMM keys, including removed eager FP8-block rows, from graph-timed measurements; all-precision kernel evidence and limits: docs/data/pr244/README.md (PR #244)." + approved_by: "simone-chen" diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/attention/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/attention/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..d159dd5880 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/attention/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,31 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + context_attention_perf: + rows: 52836 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:262799b47ef42d0dae5ced92845517d20da265daea5c58dee5813f84c7fe5ec0 + case_plan_hash: sha256:04acb1ef5df69580c3028de4dbe34f5e7697450ce783366632247395a3ed3507 + collected_at: '2026-09-16' + rows: 52836 + status: complete + generation_attention_perf: + rows: 65408 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:262799b47ef42d0dae5ced92845517d20da265daea5c58dee5813f84c7fe5ec0 + case_plan_hash: sha256:12df4833c99374841693e9bec6da73b3e7a8300acb5ca88e92c5c43d4bb24cdc + collected_at: '2026-09-16' + rows: 65408 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/attention/vllm/0.25.0/context_attention_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/attention/vllm/0.25.0/context_attention_perf.parquet new file mode 100644 index 0000000000..d4145095a9 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/attention/vllm/0.25.0/context_attention_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/attention/vllm/0.25.0/generation_attention_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/attention/vllm/0.25.0/generation_attention_perf.parquet new file mode 100644 index 0000000000..995303d4ec Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/attention/vllm/0.25.0/generation_attention_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/encoder_attention/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/encoder_attention/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..dc0d98de99 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/encoder_attention/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,21 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + encoder_attention_perf: + rows: 7679 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:b65cba616f0be2a3d04cc4192ee7e46b096784accdf6e1bb5e193987894a8830 + case_plan_hash: sha256:8a4f0e45a56a4b73354b445a9b3bccde722835cb6c6c133991182ba092eb1296 + collected_at: '2026-09-16' + rows: 7679 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/encoder_attention/vllm/0.25.0/encoder_attention_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/encoder_attention/vllm/0.25.0/encoder_attention_perf.parquet new file mode 100644 index 0000000000..2674ee3965 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/encoder_attention/vllm/0.25.0/encoder_attention_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/gemm/vllm/0.24.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/gemm/vllm/0.24.0/collection_meta.yaml index da9d3df044..95c1897814 100644 --- a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/gemm/vllm/0.24.0/collection_meta.yaml +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/gemm/vllm/0.24.0/collection_meta.yaml @@ -8,4 +8,6 @@ runtime: version: 0.24.0 tables: gemm_perf: - status: complete + status: partial + rows: 107226 + # Eager FP8-block timings removed; reuse.yaml declares the 0.25.0 donor. diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/gemm/vllm/0.24.0/gemm_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/gemm/vllm/0.24.0/gemm_perf.parquet index cf41644f09..6b8b49d0fe 100644 Binary files a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/gemm/vllm/0.24.0/gemm_perf.parquet and b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/gemm/vllm/0.24.0/gemm_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/gemm/vllm/0.24.0/reuse.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/gemm/vllm/0.24.0/reuse.yaml new file mode 100644 index 0000000000..6d5644a69f --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/gemm/vllm/0.24.0/reuse.yaml @@ -0,0 +1,9 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 1 +reuse: + - table: gemm_perf + from_version: "0.25.0" + reason: "Fill missing GEMM keys, including removed eager FP8-block rows, from graph-timed measurements; all-precision kernel evidence and limits: docs/data/pr244/README.md (PR #244)." + approved_by: "simone-chen" diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/gemm/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/gemm/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..05b0e5735f --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/gemm/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,57 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + gemm_perf: + rows: 148962 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:37d5ec63f630c7a1fc241bcb3c59e2de5e52e68b54849496160dfedf0a12fe3a + collected_at: '2026-09-16' + rows: 37518 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:a26b4b0dd1684df8b9645fddcb12b080ee89d9c39f0e08b3cb44f22926a5500e + collected_at: '2026-09-16' + rows: 37518 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:ac15c6379b8de03e49c419d74c5b702b2596c387a07d78f833da2ff288133956 + collected_at: '2026-09-16' + rows: 37518 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:e0f07c473cae107b8c2a5520c2bb1ae85045e499a268a73f0ba1ce97d035a42a + collected_at: '2026-09-16' + rows: 9348 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:fee7715156b25ee24e67995e9ae17a3bede2bbeed890c479b99cc84ade4d8fc8 + collected_at: '2026-09-16' + rows: 9348 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:03318633ff86a012c7e9bf8310f97ba545972b973af130273420a0c381c85a66 + collected_at: '2026-09-16' + rows: 8856 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:c5da28e5b245195dbf0816485f8230a16d02cf3aa39e8fed9dcffb8da85e87c8 + collected_at: '2026-09-16' + rows: 8856 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/gemm/vllm/0.25.0/gemm_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/gemm/vllm/0.25.0/gemm_perf.parquet new file mode 100644 index 0000000000..698c536362 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/gemm/vllm/0.25.0/gemm_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/linear_attention/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/linear_attention/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..4d1fbd1727 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/linear_attention/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,21 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + gdn_perf: + rows: 10326 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:ad983a508da4af6b521f77f75043cd8959537105866666aa1fbdd8da1298212f + case_plan_hash: sha256:ba375a4c3a8dbbe8b77cec5cfd0f9dafd245e18b99ace763cc738c9645962dda + collected_at: '2026-09-16' + rows: 10326 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/linear_attention/vllm/0.25.0/gdn_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/linear_attention/vllm/0.25.0/gdn_perf.parquet new file mode 100644 index 0000000000..2fac764f2b Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/linear_attention/vllm/0.25.0/gdn_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mhc/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mhc/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..8c24a09c01 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mhc/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,21 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + mhc_module_perf: + rows: 140 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:fc748ba7378714bee0e03107d3dec91f0d4659792df2cd2e31a51bdde4dd5c43 + case_plan_hash: sha256:4251dd697ea2e18fcaeed92b1a7cbdf452549a75d3a38284340d5da8502e03a1 + collected_at: '2026-09-16' + rows: 140 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mhc/vllm/0.25.0/mhc_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mhc/vllm/0.25.0/mhc_module_perf.parquet new file mode 100644 index 0000000000..3f6524eeb1 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mhc/vllm/0.25.0/mhc_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mla/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mla/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..6c4b8cbfca --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mla/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,31 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + mla_context_module_perf: + rows: 13059 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:113fa413702792e5aa95bf4ee7c12f5da9b4410e97fe1be13054acc088a6b7fb + case_plan_hash: sha256:5b1dd5428859b6622f336dbc519f3c5dadc310525ee9c11d72a17557f91c8b39 + collected_at: '2026-09-16' + rows: 13059 + status: complete + mla_generation_module_perf: + rows: 11400 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:113fa413702792e5aa95bf4ee7c12f5da9b4410e97fe1be13054acc088a6b7fb + case_plan_hash: sha256:3473224f524c629652b2a402f358af23f7f8a1cc2660955225ad5572182ae33e + collected_at: '2026-09-16' + rows: 11400 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mla/vllm/0.25.0/mla_context_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mla/vllm/0.25.0/mla_context_module_perf.parquet new file mode 100644 index 0000000000..72ca958e48 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mla/vllm/0.25.0/mla_context_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mla/vllm/0.25.0/mla_generation_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mla/vllm/0.25.0/mla_generation_module_perf.parquet new file mode 100644 index 0000000000..c51508511a Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mla/vllm/0.25.0/mla_generation_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mla_bmm/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mla_bmm/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..dd045cdc7d --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mla_bmm/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,27 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + mla_bmm_perf: + rows: 636 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:4c2fd32c623c9841221e74bf76389706de4171beb0c22793246b0ce7ea79af87 + case_plan_hash: sha256:3928a5f969ae055ba17244c0d53b504e8253870ffcbf60696731071ddb98e1e5 + collected_at: '2026-09-16' + rows: 300 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:4c2fd32c623c9841221e74bf76389706de4171beb0c22793246b0ce7ea79af87 + case_plan_hash: sha256:faeaa20a84179b59df5283aa9a238572a431f4f16f8521470519147073fe4b53 + collected_at: '2026-09-16' + rows: 336 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mla_bmm/vllm/0.25.0/mla_bmm_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mla_bmm/vllm/0.25.0/mla_bmm_perf.parquet new file mode 100644 index 0000000000..ff572dc459 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/mla_bmm/vllm/0.25.0/mla_bmm_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/moe/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/moe/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..89dab20b67 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/moe/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,27 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + moe_perf: + rows: 72924 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:fe2d2d23d340ded7bff27c17b5682e950458c2946d84d1a04bf8608cdb5a689d + case_plan_hash: sha256:261fd9888e2e292a179050ceb03bf8f7ab9d4f27ccd033f5075a92c165aec537 + collected_at: '2026-09-16' + rows: 72559 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:fe2d2d23d340ded7bff27c17b5682e950458c2946d84d1a04bf8608cdb5a689d + case_plan_hash: sha256:999bf157086b0c2d5cad715cea1ca4587a76130bad8adf97ef81b52b94636387 + collected_at: '2026-09-16' + rows: 402 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/moe/vllm/0.25.0/moe_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/moe/vllm/0.25.0/moe_perf.parquet new file mode 100644 index 0000000000..2b850399e0 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/moe/vllm/0.25.0/moe_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/msa/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/msa/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..0325862123 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/msa/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,27 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + msa_context_module_perf: + rows: 14137 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:31c1c4310a284442a8275ef4c55c51eb54ff9ba934f18f3ceb9e2b5d002f2623 + case_plan_hash: sha256:a887b4c5331833a3ac788bdfa89e15d536beaa52bb4794a0fd27e94d05a17e47 + collected_at: '2026-09-16' + rows: 14093 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:31c1c4310a284442a8275ef4c55c51eb54ff9ba934f18f3ceb9e2b5d002f2623 + case_plan_hash: sha256:a020d6ba3e1335800443fb6fc09fe930a8c274f3b6d362993b66262a012824df + collected_at: '2026-09-16' + rows: 44 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/msa/vllm/0.25.0/msa_context_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/msa/vllm/0.25.0/msa_context_module_perf.parquet new file mode 100644 index 0000000000..cd7e77ba97 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/msa/vllm/0.25.0/msa_context_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..890fb99228 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,71 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + dsv4_csa_context_module_perf: + rows: 5840 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:38a6cd429db806b370f13a9b847ad836440166e29d1e68c8e0c56da13396d5db + collected_at: '2026-09-16' + rows: 5840 + status: complete + dsv4_csa_generation_module_perf: + rows: 1544 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:21f9356813701f48d6451510032ec2746c9a64e3433c7f9954a0156c187dcae0 + collected_at: '2026-09-16' + rows: 1544 + status: complete + dsv4_hca_attn_module_perf: + rows: 3918 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:0e844f577823301a0af954d0d20baab4642bb1c0368652998bc28160dfab816d + collected_at: '2026-09-16' + rows: 3918 + status: complete + dsv4_hca_context_module_perf: + rows: 5840 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:d36420598033f22968df6d517a051324bdac1f448ff41be9aacf12beec3dd893 + collected_at: '2026-09-16' + rows: 5840 + status: complete + dsv4_hca_generation_module_perf: + rows: 1544 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:f6f9fe724d2fa5ad60fb6d0a162f93cd7901bd26093925a5fef0cbae9f310eb4 + collected_at: '2026-09-16' + rows: 1544 + status: complete + dsv4_paged_mqa_logits_module_perf: + rows: 3918 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:0b35d1427f37f58a74ae69d6503f0eecb1ba5d60499325b10e1dad10a8199250 + collected_at: '2026-09-16' + rows: 3918 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_context_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_context_module_perf.parquet new file mode 100644 index 0000000000..94776b41c8 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_context_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_generation_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_generation_module_perf.parquet new file mode 100644 index 0000000000..9d6eaec9bb Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_generation_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_attn_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_attn_module_perf.parquet new file mode 100644 index 0000000000..59d4ba25b4 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_attn_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_context_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_context_module_perf.parquet new file mode 100644 index 0000000000..d174569bc0 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_context_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_generation_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_generation_module_perf.parquet new file mode 100644 index 0000000000..26e2aec723 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_generation_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_paged_mqa_logits_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_paged_mqa_logits_module_perf.parquet new file mode 100644 index 0000000000..e4cbf8ddeb Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/b300_sxm/sparse_attention/vllm/0.25.0/dsv4_paged_mqa_logits_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/attention/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/attention/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..7cac5d161d --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/attention/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,31 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:2f726db9babd627fb59addaed2038577fc767108680901271b069d91759d1286 +tables: + context_attention_perf: + rows: 52836 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:262799b47ef42d0dae5ced92845517d20da265daea5c58dee5813f84c7fe5ec0 + case_plan_hash: sha256:04acb1ef5df69580c3028de4dbe34f5e7697450ce783366632247395a3ed3507 + collected_at: '2026-09-17' + rows: 52836 + status: complete + generation_attention_perf: + rows: 65408 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:262799b47ef42d0dae5ced92845517d20da265daea5c58dee5813f84c7fe5ec0 + case_plan_hash: sha256:12df4833c99374841693e9bec6da73b3e7a8300acb5ca88e92c5c43d4bb24cdc + collected_at: '2026-09-17' + rows: 65408 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/attention/vllm/0.25.0/context_attention_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/attention/vllm/0.25.0/context_attention_perf.parquet new file mode 100644 index 0000000000..498f1fcb7d Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/attention/vllm/0.25.0/context_attention_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/attention/vllm/0.25.0/generation_attention_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/attention/vllm/0.25.0/generation_attention_perf.parquet new file mode 100644 index 0000000000..0ef7ad43fb Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/attention/vllm/0.25.0/generation_attention_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/encoder_attention/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/encoder_attention/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..3dbf5c9c70 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/encoder_attention/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,21 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:2f726db9babd627fb59addaed2038577fc767108680901271b069d91759d1286 +tables: + encoder_attention_perf: + rows: 7679 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:b65cba616f0be2a3d04cc4192ee7e46b096784accdf6e1bb5e193987894a8830 + case_plan_hash: sha256:8a4f0e45a56a4b73354b445a9b3bccde722835cb6c6c133991182ba092eb1296 + collected_at: '2026-09-17' + rows: 7679 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/encoder_attention/vllm/0.25.0/encoder_attention_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/encoder_attention/vllm/0.25.0/encoder_attention_perf.parquet new file mode 100644 index 0000000000..1466044273 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/encoder_attention/vllm/0.25.0/encoder_attention_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/gemm/vllm/0.24.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/gemm/vllm/0.24.0/collection_meta.yaml index da9d3df044..95c1897814 100644 --- a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/gemm/vllm/0.24.0/collection_meta.yaml +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/gemm/vllm/0.24.0/collection_meta.yaml @@ -8,4 +8,6 @@ runtime: version: 0.24.0 tables: gemm_perf: - status: complete + status: partial + rows: 107226 + # Eager FP8-block timings removed; reuse.yaml declares the 0.25.0 donor. diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/gemm/vllm/0.24.0/gemm_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/gemm/vllm/0.24.0/gemm_perf.parquet index 286913e05a..7cc1a8a89c 100644 Binary files a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/gemm/vllm/0.24.0/gemm_perf.parquet and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/gemm/vllm/0.24.0/gemm_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/gemm/vllm/0.24.0/reuse.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/gemm/vllm/0.24.0/reuse.yaml new file mode 100644 index 0000000000..6d5644a69f --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/gemm/vllm/0.24.0/reuse.yaml @@ -0,0 +1,9 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 1 +reuse: + - table: gemm_perf + from_version: "0.25.0" + reason: "Fill missing GEMM keys, including removed eager FP8-block rows, from graph-timed measurements; all-precision kernel evidence and limits: docs/data/pr244/README.md (PR #244)." + approved_by: "simone-chen" diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/gemm/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/gemm/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..bf91711022 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/gemm/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,57 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:2f726db9babd627fb59addaed2038577fc767108680901271b069d91759d1286 +tables: + gemm_perf: + rows: 148962 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:37d5ec63f630c7a1fc241bcb3c59e2de5e52e68b54849496160dfedf0a12fe3a + collected_at: '2026-09-17' + rows: 37518 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:a26b4b0dd1684df8b9645fddcb12b080ee89d9c39f0e08b3cb44f22926a5500e + collected_at: '2026-09-17' + rows: 37518 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:ac15c6379b8de03e49c419d74c5b702b2596c387a07d78f833da2ff288133956 + collected_at: '2026-09-17' + rows: 37518 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:e0f07c473cae107b8c2a5520c2bb1ae85045e499a268a73f0ba1ce97d035a42a + collected_at: '2026-09-17' + rows: 9348 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:fee7715156b25ee24e67995e9ae17a3bede2bbeed890c479b99cc84ade4d8fc8 + collected_at: '2026-09-17' + rows: 9348 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:03318633ff86a012c7e9bf8310f97ba545972b973af130273420a0c381c85a66 + collected_at: '2026-09-17' + rows: 8856 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:c5da28e5b245195dbf0816485f8230a16d02cf3aa39e8fed9dcffb8da85e87c8 + collected_at: '2026-09-17' + rows: 8856 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/gemm/vllm/0.25.0/gemm_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/gemm/vllm/0.25.0/gemm_perf.parquet new file mode 100644 index 0000000000..8b0c7b08f9 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/gemm/vllm/0.25.0/gemm_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/linear_attention/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/linear_attention/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..a7794bbbea --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/linear_attention/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,21 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:2f726db9babd627fb59addaed2038577fc767108680901271b069d91759d1286 +tables: + gdn_perf: + rows: 10326 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:ad983a508da4af6b521f77f75043cd8959537105866666aa1fbdd8da1298212f + case_plan_hash: sha256:ba375a4c3a8dbbe8b77cec5cfd0f9dafd245e18b99ace763cc738c9645962dda + collected_at: '2026-09-17' + rows: 10326 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/linear_attention/vllm/0.25.0/gdn_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/linear_attention/vllm/0.25.0/gdn_perf.parquet new file mode 100644 index 0000000000..d5697058e7 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/linear_attention/vllm/0.25.0/gdn_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mhc/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mhc/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..b810f6b3ee --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mhc/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,21 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:2f726db9babd627fb59addaed2038577fc767108680901271b069d91759d1286 +tables: + mhc_module_perf: + rows: 140 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:fc748ba7378714bee0e03107d3dec91f0d4659792df2cd2e31a51bdde4dd5c43 + case_plan_hash: sha256:4251dd697ea2e18fcaeed92b1a7cbdf452549a75d3a38284340d5da8502e03a1 + collected_at: '2026-09-17' + rows: 140 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mhc/vllm/0.25.0/mhc_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mhc/vllm/0.25.0/mhc_module_perf.parquet new file mode 100644 index 0000000000..c151d1fd85 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mhc/vllm/0.25.0/mhc_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mla/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mla/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..a2e7239f67 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mla/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,31 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:2f726db9babd627fb59addaed2038577fc767108680901271b069d91759d1286 +tables: + mla_context_module_perf: + rows: 13059 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:113fa413702792e5aa95bf4ee7c12f5da9b4410e97fe1be13054acc088a6b7fb + case_plan_hash: sha256:5b1dd5428859b6622f336dbc519f3c5dadc310525ee9c11d72a17557f91c8b39 + collected_at: '2026-09-17' + rows: 13059 + status: complete + mla_generation_module_perf: + rows: 11400 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:113fa413702792e5aa95bf4ee7c12f5da9b4410e97fe1be13054acc088a6b7fb + case_plan_hash: sha256:3473224f524c629652b2a402f358af23f7f8a1cc2660955225ad5572182ae33e + collected_at: '2026-09-17' + rows: 11400 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mla/vllm/0.25.0/mla_context_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mla/vllm/0.25.0/mla_context_module_perf.parquet new file mode 100644 index 0000000000..c13f5e901e Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mla/vllm/0.25.0/mla_context_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mla/vllm/0.25.0/mla_generation_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mla/vllm/0.25.0/mla_generation_module_perf.parquet new file mode 100644 index 0000000000..16a9bdf80a Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mla/vllm/0.25.0/mla_generation_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mla_bmm/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mla_bmm/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..16ea7626a0 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mla_bmm/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,27 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:2f726db9babd627fb59addaed2038577fc767108680901271b069d91759d1286 +tables: + mla_bmm_perf: + rows: 636 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:4c2fd32c623c9841221e74bf76389706de4171beb0c22793246b0ce7ea79af87 + case_plan_hash: sha256:3928a5f969ae055ba17244c0d53b504e8253870ffcbf60696731071ddb98e1e5 + collected_at: '2026-09-17' + rows: 300 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:4c2fd32c623c9841221e74bf76389706de4171beb0c22793246b0ce7ea79af87 + case_plan_hash: sha256:faeaa20a84179b59df5283aa9a238572a431f4f16f8521470519147073fe4b53 + collected_at: '2026-09-17' + rows: 336 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mla_bmm/vllm/0.25.0/mla_bmm_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mla_bmm/vllm/0.25.0/mla_bmm_perf.parquet new file mode 100644 index 0000000000..705128b973 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/mla_bmm/vllm/0.25.0/mla_bmm_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/moe/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/moe/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..f4814742a5 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/moe/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,27 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:2f726db9babd627fb59addaed2038577fc767108680901271b069d91759d1286 +tables: + moe_perf: + rows: 72976 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:fe2d2d23d340ded7bff27c17b5682e950458c2946d84d1a04bf8608cdb5a689d + case_plan_hash: sha256:261fd9888e2e292a179050ceb03bf8f7ab9d4f27ccd033f5075a92c165aec537 + collected_at: '2026-09-17' + rows: 72479 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:fe2d2d23d340ded7bff27c17b5682e950458c2946d84d1a04bf8608cdb5a689d + case_plan_hash: sha256:0d641f020b07c0338787cbaf71f489d8ed0290dbb052d0de69dc1084bab098ec + collected_at: '2026-09-17' + rows: 535 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/moe/vllm/0.25.0/moe_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/moe/vllm/0.25.0/moe_perf.parquet new file mode 100644 index 0000000000..6ae4748f7a Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/moe/vllm/0.25.0/moe_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/msa/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/msa/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..601fc76bf5 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/msa/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,27 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:2f726db9babd627fb59addaed2038577fc767108680901271b069d91759d1286 +tables: + msa_context_module_perf: + rows: 14108 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:31c1c4310a284442a8275ef4c55c51eb54ff9ba934f18f3ceb9e2b5d002f2623 + case_plan_hash: sha256:a887b4c5331833a3ac788bdfa89e15d536beaa52bb4794a0fd27e94d05a17e47 + collected_at: '2026-09-17' + rows: 14060 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:31c1c4310a284442a8275ef4c55c51eb54ff9ba934f18f3ceb9e2b5d002f2623 + case_plan_hash: sha256:62829916eb427cf123603c1545a90f21c38489c26ee678e6932106a28ce29828 + collected_at: '2026-09-17' + rows: 48 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/msa/vllm/0.25.0/msa_context_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/msa/vllm/0.25.0/msa_context_module_perf.parquet new file mode 100644 index 0000000000..efecd31880 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/msa/vllm/0.25.0/msa_context_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..df90df3d9c --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,71 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:2f726db9babd627fb59addaed2038577fc767108680901271b069d91759d1286 +tables: + dsv4_csa_context_module_perf: + rows: 5840 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:38a6cd429db806b370f13a9b847ad836440166e29d1e68c8e0c56da13396d5db + collected_at: '2026-09-17' + rows: 5840 + status: complete + dsv4_csa_generation_module_perf: + rows: 1544 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:21f9356813701f48d6451510032ec2746c9a64e3433c7f9954a0156c187dcae0 + collected_at: '2026-09-17' + rows: 1544 + status: complete + dsv4_hca_attn_module_perf: + rows: 3918 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:0e844f577823301a0af954d0d20baab4642bb1c0368652998bc28160dfab816d + collected_at: '2026-09-17' + rows: 3918 + status: complete + dsv4_hca_context_module_perf: + rows: 5840 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:d36420598033f22968df6d517a051324bdac1f448ff41be9aacf12beec3dd893 + collected_at: '2026-09-17' + rows: 5840 + status: complete + dsv4_hca_generation_module_perf: + rows: 1544 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:f6f9fe724d2fa5ad60fb6d0a162f93cd7901bd26093925a5fef0cbae9f310eb4 + collected_at: '2026-09-17' + rows: 1544 + status: complete + dsv4_paged_mqa_logits_module_perf: + rows: 3918 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:0b35d1427f37f58a74ae69d6503f0eecb1ba5d60499325b10e1dad10a8199250 + collected_at: '2026-09-17' + rows: 3918 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_csa_context_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_csa_context_module_perf.parquet new file mode 100644 index 0000000000..443bb9e501 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_csa_context_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_csa_generation_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_csa_generation_module_perf.parquet new file mode 100644 index 0000000000..fb6821c188 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_csa_generation_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_hca_attn_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_hca_attn_module_perf.parquet new file mode 100644 index 0000000000..a57023a83d Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_hca_attn_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_hca_context_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_hca_context_module_perf.parquet new file mode 100644 index 0000000000..81606a322f Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_hca_context_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_hca_generation_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_hca_generation_module_perf.parquet new file mode 100644 index 0000000000..424351bc8d Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_hca_generation_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_paged_mqa_logits_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_paged_mqa_logits_module_perf.parquet new file mode 100644 index 0000000000..b2b849de0e Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb200/sparse_attention/vllm/0.25.0/dsv4_paged_mqa_logits_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/attention/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/attention/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..7cac5d161d --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/attention/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,31 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:2f726db9babd627fb59addaed2038577fc767108680901271b069d91759d1286 +tables: + context_attention_perf: + rows: 52836 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:262799b47ef42d0dae5ced92845517d20da265daea5c58dee5813f84c7fe5ec0 + case_plan_hash: sha256:04acb1ef5df69580c3028de4dbe34f5e7697450ce783366632247395a3ed3507 + collected_at: '2026-09-17' + rows: 52836 + status: complete + generation_attention_perf: + rows: 65408 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:262799b47ef42d0dae5ced92845517d20da265daea5c58dee5813f84c7fe5ec0 + case_plan_hash: sha256:12df4833c99374841693e9bec6da73b3e7a8300acb5ca88e92c5c43d4bb24cdc + collected_at: '2026-09-17' + rows: 65408 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/attention/vllm/0.25.0/context_attention_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/attention/vllm/0.25.0/context_attention_perf.parquet new file mode 100644 index 0000000000..9b37fc5223 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/attention/vllm/0.25.0/context_attention_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/attention/vllm/0.25.0/generation_attention_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/attention/vllm/0.25.0/generation_attention_perf.parquet new file mode 100644 index 0000000000..ac03f151bd Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/attention/vllm/0.25.0/generation_attention_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/encoder_attention/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/encoder_attention/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..3dbf5c9c70 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/encoder_attention/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,21 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:2f726db9babd627fb59addaed2038577fc767108680901271b069d91759d1286 +tables: + encoder_attention_perf: + rows: 7679 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:b65cba616f0be2a3d04cc4192ee7e46b096784accdf6e1bb5e193987894a8830 + case_plan_hash: sha256:8a4f0e45a56a4b73354b445a9b3bccde722835cb6c6c133991182ba092eb1296 + collected_at: '2026-09-17' + rows: 7679 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/encoder_attention/vllm/0.25.0/encoder_attention_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/encoder_attention/vllm/0.25.0/encoder_attention_perf.parquet new file mode 100644 index 0000000000..62b7d63e0f Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/encoder_attention/vllm/0.25.0/encoder_attention_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/gemm/vllm/0.24.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/gemm/vllm/0.24.0/collection_meta.yaml index da9d3df044..95c1897814 100644 --- a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/gemm/vllm/0.24.0/collection_meta.yaml +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/gemm/vllm/0.24.0/collection_meta.yaml @@ -8,4 +8,6 @@ runtime: version: 0.24.0 tables: gemm_perf: - status: complete + status: partial + rows: 107226 + # Eager FP8-block timings removed; reuse.yaml declares the 0.25.0 donor. diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/gemm/vllm/0.24.0/gemm_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/gemm/vllm/0.24.0/gemm_perf.parquet index 1cccfc33ba..9801bada6e 100644 Binary files a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/gemm/vllm/0.24.0/gemm_perf.parquet and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/gemm/vllm/0.24.0/gemm_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/gemm/vllm/0.24.0/reuse.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/gemm/vllm/0.24.0/reuse.yaml new file mode 100644 index 0000000000..6d5644a69f --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/gemm/vllm/0.24.0/reuse.yaml @@ -0,0 +1,9 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 1 +reuse: + - table: gemm_perf + from_version: "0.25.0" + reason: "Fill missing GEMM keys, including removed eager FP8-block rows, from graph-timed measurements; all-precision kernel evidence and limits: docs/data/pr244/README.md (PR #244)." + approved_by: "simone-chen" diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/gemm/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/gemm/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..bf91711022 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/gemm/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,57 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:2f726db9babd627fb59addaed2038577fc767108680901271b069d91759d1286 +tables: + gemm_perf: + rows: 148962 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:37d5ec63f630c7a1fc241bcb3c59e2de5e52e68b54849496160dfedf0a12fe3a + collected_at: '2026-09-17' + rows: 37518 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:a26b4b0dd1684df8b9645fddcb12b080ee89d9c39f0e08b3cb44f22926a5500e + collected_at: '2026-09-17' + rows: 37518 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:ac15c6379b8de03e49c419d74c5b702b2596c387a07d78f833da2ff288133956 + collected_at: '2026-09-17' + rows: 37518 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:e0f07c473cae107b8c2a5520c2bb1ae85045e499a268a73f0ba1ce97d035a42a + collected_at: '2026-09-17' + rows: 9348 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:fee7715156b25ee24e67995e9ae17a3bede2bbeed890c479b99cc84ade4d8fc8 + collected_at: '2026-09-17' + rows: 9348 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:03318633ff86a012c7e9bf8310f97ba545972b973af130273420a0c381c85a66 + collected_at: '2026-09-17' + rows: 8856 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:c5da28e5b245195dbf0816485f8230a16d02cf3aa39e8fed9dcffb8da85e87c8 + collected_at: '2026-09-17' + rows: 8856 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/gemm/vllm/0.25.0/gemm_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/gemm/vllm/0.25.0/gemm_perf.parquet new file mode 100644 index 0000000000..27a423d513 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/gemm/vllm/0.25.0/gemm_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/linear_attention/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/linear_attention/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..a7794bbbea --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/linear_attention/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,21 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:2f726db9babd627fb59addaed2038577fc767108680901271b069d91759d1286 +tables: + gdn_perf: + rows: 10326 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:ad983a508da4af6b521f77f75043cd8959537105866666aa1fbdd8da1298212f + case_plan_hash: sha256:ba375a4c3a8dbbe8b77cec5cfd0f9dafd245e18b99ace763cc738c9645962dda + collected_at: '2026-09-17' + rows: 10326 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/linear_attention/vllm/0.25.0/gdn_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/linear_attention/vllm/0.25.0/gdn_perf.parquet new file mode 100644 index 0000000000..fe74a9d3bb Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/linear_attention/vllm/0.25.0/gdn_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mhc/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mhc/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..b810f6b3ee --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mhc/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,21 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:2f726db9babd627fb59addaed2038577fc767108680901271b069d91759d1286 +tables: + mhc_module_perf: + rows: 140 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:fc748ba7378714bee0e03107d3dec91f0d4659792df2cd2e31a51bdde4dd5c43 + case_plan_hash: sha256:4251dd697ea2e18fcaeed92b1a7cbdf452549a75d3a38284340d5da8502e03a1 + collected_at: '2026-09-17' + rows: 140 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mhc/vllm/0.25.0/mhc_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mhc/vllm/0.25.0/mhc_module_perf.parquet new file mode 100644 index 0000000000..8152bf3ea2 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mhc/vllm/0.25.0/mhc_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mla/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mla/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..a2e7239f67 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mla/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,31 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:2f726db9babd627fb59addaed2038577fc767108680901271b069d91759d1286 +tables: + mla_context_module_perf: + rows: 13059 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:113fa413702792e5aa95bf4ee7c12f5da9b4410e97fe1be13054acc088a6b7fb + case_plan_hash: sha256:5b1dd5428859b6622f336dbc519f3c5dadc310525ee9c11d72a17557f91c8b39 + collected_at: '2026-09-17' + rows: 13059 + status: complete + mla_generation_module_perf: + rows: 11400 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:113fa413702792e5aa95bf4ee7c12f5da9b4410e97fe1be13054acc088a6b7fb + case_plan_hash: sha256:3473224f524c629652b2a402f358af23f7f8a1cc2660955225ad5572182ae33e + collected_at: '2026-09-17' + rows: 11400 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mla/vllm/0.25.0/mla_context_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mla/vllm/0.25.0/mla_context_module_perf.parquet new file mode 100644 index 0000000000..9829ce54a2 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mla/vllm/0.25.0/mla_context_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mla/vllm/0.25.0/mla_generation_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mla/vllm/0.25.0/mla_generation_module_perf.parquet new file mode 100644 index 0000000000..75f0061393 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mla/vllm/0.25.0/mla_generation_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mla_bmm/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mla_bmm/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..16ea7626a0 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mla_bmm/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,27 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:2f726db9babd627fb59addaed2038577fc767108680901271b069d91759d1286 +tables: + mla_bmm_perf: + rows: 636 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:4c2fd32c623c9841221e74bf76389706de4171beb0c22793246b0ce7ea79af87 + case_plan_hash: sha256:3928a5f969ae055ba17244c0d53b504e8253870ffcbf60696731071ddb98e1e5 + collected_at: '2026-09-17' + rows: 300 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:4c2fd32c623c9841221e74bf76389706de4171beb0c22793246b0ce7ea79af87 + case_plan_hash: sha256:faeaa20a84179b59df5283aa9a238572a431f4f16f8521470519147073fe4b53 + collected_at: '2026-09-17' + rows: 336 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mla_bmm/vllm/0.25.0/mla_bmm_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mla_bmm/vllm/0.25.0/mla_bmm_perf.parquet new file mode 100644 index 0000000000..a7fb82fb91 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/mla_bmm/vllm/0.25.0/mla_bmm_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/moe/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/moe/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..575b365de9 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/moe/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,27 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:2f726db9babd627fb59addaed2038577fc767108680901271b069d91759d1286 +tables: + moe_perf: + rows: 73002 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:fe2d2d23d340ded7bff27c17b5682e950458c2946d84d1a04bf8608cdb5a689d + case_plan_hash: sha256:261fd9888e2e292a179050ceb03bf8f7ab9d4f27ccd033f5075a92c165aec537 + collected_at: '2026-09-17' + rows: 72458 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:fe2d2d23d340ded7bff27c17b5682e950458c2946d84d1a04bf8608cdb5a689d + case_plan_hash: sha256:5e784d6b29caea89ab2b704fcefd1538315790fa15eabf7c4c12a1ab124c6308 + collected_at: '2026-09-17' + rows: 588 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/moe/vllm/0.25.0/moe_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/moe/vllm/0.25.0/moe_perf.parquet new file mode 100644 index 0000000000..227f12e7d1 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/moe/vllm/0.25.0/moe_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/msa/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/msa/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..248a72d63f --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/msa/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,27 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:2f726db9babd627fb59addaed2038577fc767108680901271b069d91759d1286 +tables: + msa_context_module_perf: + rows: 14174 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:31c1c4310a284442a8275ef4c55c51eb54ff9ba934f18f3ceb9e2b5d002f2623 + case_plan_hash: sha256:a887b4c5331833a3ac788bdfa89e15d536beaa52bb4794a0fd27e94d05a17e47 + collected_at: '2026-09-17' + rows: 14123 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:31c1c4310a284442a8275ef4c55c51eb54ff9ba934f18f3ceb9e2b5d002f2623 + case_plan_hash: sha256:f9f4ea0aa7d4344a312c351ea0a4b2120590e0563571c7d442583c273efc972b + collected_at: '2026-09-17' + rows: 51 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/msa/vllm/0.25.0/msa_context_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/msa/vllm/0.25.0/msa_context_module_perf.parquet new file mode 100644 index 0000000000..6225a8fac4 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/msa/vllm/0.25.0/msa_context_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..df90df3d9c --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,71 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:2f726db9babd627fb59addaed2038577fc767108680901271b069d91759d1286 +tables: + dsv4_csa_context_module_perf: + rows: 5840 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:38a6cd429db806b370f13a9b847ad836440166e29d1e68c8e0c56da13396d5db + collected_at: '2026-09-17' + rows: 5840 + status: complete + dsv4_csa_generation_module_perf: + rows: 1544 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:21f9356813701f48d6451510032ec2746c9a64e3433c7f9954a0156c187dcae0 + collected_at: '2026-09-17' + rows: 1544 + status: complete + dsv4_hca_attn_module_perf: + rows: 3918 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:0e844f577823301a0af954d0d20baab4642bb1c0368652998bc28160dfab816d + collected_at: '2026-09-17' + rows: 3918 + status: complete + dsv4_hca_context_module_perf: + rows: 5840 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:d36420598033f22968df6d517a051324bdac1f448ff41be9aacf12beec3dd893 + collected_at: '2026-09-17' + rows: 5840 + status: complete + dsv4_hca_generation_module_perf: + rows: 1544 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:f6f9fe724d2fa5ad60fb6d0a162f93cd7901bd26093925a5fef0cbae9f310eb4 + collected_at: '2026-09-17' + rows: 1544 + status: complete + dsv4_paged_mqa_logits_module_perf: + rows: 3918 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:0b35d1427f37f58a74ae69d6503f0eecb1ba5d60499325b10e1dad10a8199250 + collected_at: '2026-09-17' + rows: 3918 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_csa_context_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_csa_context_module_perf.parquet new file mode 100644 index 0000000000..b10bc1e56f Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_csa_context_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_csa_generation_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_csa_generation_module_perf.parquet new file mode 100644 index 0000000000..ba671d2720 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_csa_generation_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_hca_attn_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_hca_attn_module_perf.parquet new file mode 100644 index 0000000000..f04505c645 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_hca_attn_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_hca_context_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_hca_context_module_perf.parquet new file mode 100644 index 0000000000..1a0008d45b Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_hca_context_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_hca_generation_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_hca_generation_module_perf.parquet new file mode 100644 index 0000000000..74fefde122 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_hca_generation_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_paged_mqa_logits_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_paged_mqa_logits_module_perf.parquet new file mode 100644 index 0000000000..94af65b3c0 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/gb300/sparse_attention/vllm/0.25.0/dsv4_paged_mqa_logits_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/attention/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/attention/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..1c63025efa --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/attention/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,31 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + context_attention_perf: + rows: 53771 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:262799b47ef42d0dae5ced92845517d20da265daea5c58dee5813f84c7fe5ec0 + case_plan_hash: sha256:04acb1ef5df69580c3028de4dbe34f5e7697450ce783366632247395a3ed3507 + collected_at: '2026-09-17' + rows: 53771 + status: complete + generation_attention_perf: + rows: 66708 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:262799b47ef42d0dae5ced92845517d20da265daea5c58dee5813f84c7fe5ec0 + case_plan_hash: sha256:12df4833c99374841693e9bec6da73b3e7a8300acb5ca88e92c5c43d4bb24cdc + collected_at: '2026-09-17' + rows: 66708 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/attention/vllm/0.25.0/context_attention_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/attention/vllm/0.25.0/context_attention_perf.parquet new file mode 100644 index 0000000000..71f9cca2b8 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/attention/vllm/0.25.0/context_attention_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/attention/vllm/0.25.0/generation_attention_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/attention/vllm/0.25.0/generation_attention_perf.parquet new file mode 100644 index 0000000000..db5c79491d Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/attention/vllm/0.25.0/generation_attention_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/encoder_attention/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/encoder_attention/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..afb6632e06 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/encoder_attention/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,21 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + encoder_attention_perf: + rows: 7679 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:b65cba616f0be2a3d04cc4192ee7e46b096784accdf6e1bb5e193987894a8830 + case_plan_hash: sha256:8a4f0e45a56a4b73354b445a9b3bccde722835cb6c6c133991182ba092eb1296 + collected_at: '2026-09-17' + rows: 7679 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/encoder_attention/vllm/0.25.0/encoder_attention_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/encoder_attention/vllm/0.25.0/encoder_attention_perf.parquet new file mode 100644 index 0000000000..19793d4703 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/encoder_attention/vllm/0.25.0/encoder_attention_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/gemm/vllm/0.24.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/gemm/vllm/0.24.0/collection_meta.yaml index da9d3df044..6b06ccc31a 100644 --- a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/gemm/vllm/0.24.0/collection_meta.yaml +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/gemm/vllm/0.24.0/collection_meta.yaml @@ -8,4 +8,6 @@ runtime: version: 0.24.0 tables: gemm_perf: - status: complete + status: partial + rows: 71484 + # Eager FP8-block timings removed; reuse.yaml declares the 0.25.0 donor. diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/gemm/vllm/0.24.0/gemm_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/gemm/vllm/0.24.0/gemm_perf.parquet index a5225c8185..792e46c5b4 100644 Binary files a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/gemm/vllm/0.24.0/gemm_perf.parquet and b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/gemm/vllm/0.24.0/gemm_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/gemm/vllm/0.24.0/reuse.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/gemm/vllm/0.24.0/reuse.yaml new file mode 100644 index 0000000000..6d5644a69f --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/gemm/vllm/0.24.0/reuse.yaml @@ -0,0 +1,9 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 1 +reuse: + - table: gemm_perf + from_version: "0.25.0" + reason: "Fill missing GEMM keys, including removed eager FP8-block rows, from graph-timed measurements; all-precision kernel evidence and limits: docs/data/pr244/README.md (PR #244)." + approved_by: "simone-chen" diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/gemm/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/gemm/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..2559869ec8 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/gemm/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,75 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + gemm_perf: + rows: 111444 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:37d5ec63f630c7a1fc241bcb3c59e2de5e52e68b54849496160dfedf0a12fe3a + collected_at: '2026-09-17' + rows: 37518 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:a26b4b0dd1684df8b9645fddcb12b080ee89d9c39f0e08b3cb44f22926a5500e + collected_at: '2026-09-17' + rows: 37518 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:e0f07c473cae107b8c2a5520c2bb1ae85045e499a268a73f0ba1ce97d035a42a + collected_at: '2026-09-17' + rows: 9340 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:e2f1ad98de9560d848f5221124a6316a39da55b5ce3edc69c7a81b2c44e5d31e + collected_at: '2026-09-17' + rows: 8 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:fee7715156b25ee24e67995e9ae17a3bede2bbeed890c479b99cc84ade4d8fc8 + collected_at: '2026-09-17' + rows: 9341 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:a2996b236bf7a38b450fe0ce7919e888aa88a8cf8b1e67512e9affe07fe9880c + collected_at: '2026-09-17' + rows: 7 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:03318633ff86a012c7e9bf8310f97ba545972b973af130273420a0c381c85a66 + collected_at: '2026-09-17' + rows: 8851 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:9c800e0e3cdaa7b78921fcc83d50830c46656c68b8b709e6cbedc9cdd857369a + collected_at: '2026-09-17' + rows: 5 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:c5da28e5b245195dbf0816485f8230a16d02cf3aa39e8fed9dcffb8da85e87c8 + collected_at: '2026-09-17' + rows: 8848 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:37705cea380e9a880fc56b1a1c76bc1ea3b5b52d4ee2b94d46aac751da14c952 + collected_at: '2026-09-17' + rows: 8 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/gemm/vllm/0.25.0/gemm_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/gemm/vllm/0.25.0/gemm_perf.parquet new file mode 100644 index 0000000000..30b4a891e2 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/gemm/vllm/0.25.0/gemm_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/linear_attention/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/linear_attention/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..7914dfd79f --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/linear_attention/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,21 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + gdn_perf: + rows: 10284 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:ad983a508da4af6b521f77f75043cd8959537105866666aa1fbdd8da1298212f + case_plan_hash: sha256:ba375a4c3a8dbbe8b77cec5cfd0f9dafd245e18b99ace763cc738c9645962dda + collected_at: '2026-09-17' + rows: 10284 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/linear_attention/vllm/0.25.0/gdn_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/linear_attention/vllm/0.25.0/gdn_perf.parquet new file mode 100644 index 0000000000..f972935a20 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/linear_attention/vllm/0.25.0/gdn_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mhc/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mhc/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..5f6ec1083d --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mhc/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,21 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + mhc_module_perf: + rows: 139 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:fc748ba7378714bee0e03107d3dec91f0d4659792df2cd2e31a51bdde4dd5c43 + case_plan_hash: sha256:4251dd697ea2e18fcaeed92b1a7cbdf452549a75d3a38284340d5da8502e03a1 + collected_at: '2026-09-17' + rows: 139 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mhc/vllm/0.25.0/mhc_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mhc/vllm/0.25.0/mhc_module_perf.parquet new file mode 100644 index 0000000000..38d9c4ffb6 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mhc/vllm/0.25.0/mhc_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mla/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mla/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..90432c9227 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mla/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,31 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + mla_context_module_perf: + rows: 5820 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:113fa413702792e5aa95bf4ee7c12f5da9b4410e97fe1be13054acc088a6b7fb + case_plan_hash: sha256:07e0ec5401fec176bb9e2ffe8d5ec19c54a2355f25ead16ff7e1960a279ca0df + collected_at: '2026-09-17' + rows: 5820 + status: complete + mla_generation_module_perf: + rows: 8760 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:113fa413702792e5aa95bf4ee7c12f5da9b4410e97fe1be13054acc088a6b7fb + case_plan_hash: sha256:266b8e03bae0501992a44a80eb57c80b2e1dee9e4aba26b673a247ec13e947e8 + collected_at: '2026-09-17' + rows: 8760 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mla/vllm/0.25.0/mla_context_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mla/vllm/0.25.0/mla_context_module_perf.parquet new file mode 100644 index 0000000000..6ccb6961f8 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mla/vllm/0.25.0/mla_context_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mla/vllm/0.25.0/mla_generation_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mla/vllm/0.25.0/mla_generation_module_perf.parquet new file mode 100644 index 0000000000..0b0460962e Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mla/vllm/0.25.0/mla_generation_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mla_bmm/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mla_bmm/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..04e02dce62 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mla_bmm/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,27 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + mla_bmm_perf: + rows: 636 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:4c2fd32c623c9841221e74bf76389706de4171beb0c22793246b0ce7ea79af87 + case_plan_hash: sha256:3928a5f969ae055ba17244c0d53b504e8253870ffcbf60696731071ddb98e1e5 + collected_at: '2026-09-17' + rows: 300 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:4c2fd32c623c9841221e74bf76389706de4171beb0c22793246b0ce7ea79af87 + case_plan_hash: sha256:faeaa20a84179b59df5283aa9a238572a431f4f16f8521470519147073fe4b53 + collected_at: '2026-09-17' + rows: 336 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mla_bmm/vllm/0.25.0/mla_bmm_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mla_bmm/vllm/0.25.0/mla_bmm_perf.parquet new file mode 100644 index 0000000000..e2bbcc98d3 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/mla_bmm/vllm/0.25.0/mla_bmm_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/moe/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/moe/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..e26681297b --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/moe/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,27 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + moe_perf: + rows: 54432 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:fe2d2d23d340ded7bff27c17b5682e950458c2946d84d1a04bf8608cdb5a689d + case_plan_hash: sha256:86632275c959e534d7bc4d5ee15cd3b7cf60a75638a57992d14529f70f5e4b61 + collected_at: '2026-09-17' + rows: 54242 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:fe2d2d23d340ded7bff27c17b5682e950458c2946d84d1a04bf8608cdb5a689d + case_plan_hash: sha256:9c8e38864a66fabe2b678e1b167cbbceeac33d14c76c1f51c9408740ef875fc3 + collected_at: '2026-09-17' + rows: 216 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/moe/vllm/0.25.0/moe_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/moe/vllm/0.25.0/moe_perf.parquet new file mode 100644 index 0000000000..60825398f4 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/moe/vllm/0.25.0/moe_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/msa/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/msa/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..f042cdb37a --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/msa/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,27 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + msa_context_module_perf: + rows: 11270 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:31c1c4310a284442a8275ef4c55c51eb54ff9ba934f18f3ceb9e2b5d002f2623 + case_plan_hash: sha256:2fa04db807438446fe3c239cafe10fc48f75b102de2146fa128bbbba094d9a90 + collected_at: '2026-09-17' + rows: 11199 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:31c1c4310a284442a8275ef4c55c51eb54ff9ba934f18f3ceb9e2b5d002f2623 + case_plan_hash: sha256:5ea4a952c31d82d67e5633112f386acaeb053279e289350d8de1296af25c092d + collected_at: '2026-09-17' + rows: 71 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/msa/vllm/0.25.0/msa_context_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/msa/vllm/0.25.0/msa_context_module_perf.parquet new file mode 100644 index 0000000000..9a3fb37c86 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/msa/vllm/0.25.0/msa_context_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..acb9dd93ed --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,77 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + dsv4_csa_context_module_perf: + rows: 5789 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:38a6cd429db806b370f13a9b847ad836440166e29d1e68c8e0c56da13396d5db + collected_at: '2026-09-17' + rows: 5780 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:302875e25b2e6174ca33afc82ce75d5a4c878b97196a0c6e0987fc8794eac29d + collected_at: '2026-09-17' + rows: 9 + status: complete + dsv4_csa_generation_module_perf: + rows: 1544 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:21f9356813701f48d6451510032ec2746c9a64e3433c7f9954a0156c187dcae0 + collected_at: '2026-09-17' + rows: 1544 + status: complete + dsv4_hca_attn_module_perf: + rows: 3918 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:0e844f577823301a0af954d0d20baab4642bb1c0368652998bc28160dfab816d + collected_at: '2026-09-17' + rows: 3918 + status: complete + dsv4_hca_context_module_perf: + rows: 5793 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:d36420598033f22968df6d517a051324bdac1f448ff41be9aacf12beec3dd893 + collected_at: '2026-09-17' + rows: 5793 + status: complete + dsv4_hca_generation_module_perf: + rows: 1544 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:f6f9fe724d2fa5ad60fb6d0a162f93cd7901bd26093925a5fef0cbae9f310eb4 + collected_at: '2026-09-17' + rows: 1544 + status: complete + dsv4_paged_mqa_logits_module_perf: + rows: 3918 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:0b35d1427f37f58a74ae69d6503f0eecb1ba5d60499325b10e1dad10a8199250 + collected_at: '2026-09-17' + rows: 3918 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_context_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_context_module_perf.parquet new file mode 100644 index 0000000000..0b331c23f1 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_context_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_generation_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_generation_module_perf.parquet new file mode 100644 index 0000000000..479ad4e359 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_generation_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_attn_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_attn_module_perf.parquet new file mode 100644 index 0000000000..b105008d6a Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_attn_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_context_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_context_module_perf.parquet new file mode 100644 index 0000000000..2c1af741c0 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_context_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_generation_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_generation_module_perf.parquet new file mode 100644 index 0000000000..702b55118c Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_generation_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_paged_mqa_logits_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_paged_mqa_logits_module_perf.parquet new file mode 100644 index 0000000000..691c0f57ea Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h100_sxm/sparse_attention/vllm/0.25.0/dsv4_paged_mqa_logits_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/attention/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/attention/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..1c63025efa --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/attention/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,31 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + context_attention_perf: + rows: 53771 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:262799b47ef42d0dae5ced92845517d20da265daea5c58dee5813f84c7fe5ec0 + case_plan_hash: sha256:04acb1ef5df69580c3028de4dbe34f5e7697450ce783366632247395a3ed3507 + collected_at: '2026-09-17' + rows: 53771 + status: complete + generation_attention_perf: + rows: 66708 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:262799b47ef42d0dae5ced92845517d20da265daea5c58dee5813f84c7fe5ec0 + case_plan_hash: sha256:12df4833c99374841693e9bec6da73b3e7a8300acb5ca88e92c5c43d4bb24cdc + collected_at: '2026-09-17' + rows: 66708 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/attention/vllm/0.25.0/context_attention_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/attention/vllm/0.25.0/context_attention_perf.parquet new file mode 100644 index 0000000000..711036b6eb Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/attention/vllm/0.25.0/context_attention_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/attention/vllm/0.25.0/generation_attention_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/attention/vllm/0.25.0/generation_attention_perf.parquet new file mode 100644 index 0000000000..ac8a5e3600 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/attention/vllm/0.25.0/generation_attention_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/encoder_attention/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/encoder_attention/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..afb6632e06 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/encoder_attention/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,21 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + encoder_attention_perf: + rows: 7679 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:b65cba616f0be2a3d04cc4192ee7e46b096784accdf6e1bb5e193987894a8830 + case_plan_hash: sha256:8a4f0e45a56a4b73354b445a9b3bccde722835cb6c6c133991182ba092eb1296 + collected_at: '2026-09-17' + rows: 7679 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/encoder_attention/vllm/0.25.0/encoder_attention_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/encoder_attention/vllm/0.25.0/encoder_attention_perf.parquet new file mode 100644 index 0000000000..3f5a4622eb Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/encoder_attention/vllm/0.25.0/encoder_attention_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/gemm/vllm/0.24.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/gemm/vllm/0.24.0/collection_meta.yaml index da9d3df044..6b06ccc31a 100644 --- a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/gemm/vllm/0.24.0/collection_meta.yaml +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/gemm/vllm/0.24.0/collection_meta.yaml @@ -8,4 +8,6 @@ runtime: version: 0.24.0 tables: gemm_perf: - status: complete + status: partial + rows: 71484 + # Eager FP8-block timings removed; reuse.yaml declares the 0.25.0 donor. diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/gemm/vllm/0.24.0/gemm_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/gemm/vllm/0.24.0/gemm_perf.parquet index ebd4c8faca..cc2d1fad20 100644 Binary files a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/gemm/vllm/0.24.0/gemm_perf.parquet and b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/gemm/vllm/0.24.0/gemm_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/gemm/vllm/0.24.0/reuse.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/gemm/vllm/0.24.0/reuse.yaml new file mode 100644 index 0000000000..6d5644a69f --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/gemm/vllm/0.24.0/reuse.yaml @@ -0,0 +1,9 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 1 +reuse: + - table: gemm_perf + from_version: "0.25.0" + reason: "Fill missing GEMM keys, including removed eager FP8-block rows, from graph-timed measurements; all-precision kernel evidence and limits: docs/data/pr244/README.md (PR #244)." + approved_by: "simone-chen" diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/gemm/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/gemm/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..e4e0ea7ddc --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/gemm/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,51 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + gemm_perf: + rows: 111444 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:37d5ec63f630c7a1fc241bcb3c59e2de5e52e68b54849496160dfedf0a12fe3a + collected_at: '2026-09-17' + rows: 37518 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:a26b4b0dd1684df8b9645fddcb12b080ee89d9c39f0e08b3cb44f22926a5500e + collected_at: '2026-09-17' + rows: 37518 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:e0f07c473cae107b8c2a5520c2bb1ae85045e499a268a73f0ba1ce97d035a42a + collected_at: '2026-09-17' + rows: 9348 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:fee7715156b25ee24e67995e9ae17a3bede2bbeed890c479b99cc84ade4d8fc8 + collected_at: '2026-09-17' + rows: 9348 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:03318633ff86a012c7e9bf8310f97ba545972b973af130273420a0c381c85a66 + collected_at: '2026-09-17' + rows: 8856 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5d490b4e88354bf5d27eb98a392bd4b951d9a2331f093d40e51c79d32c0365d5 + case_plan_hash: sha256:c5da28e5b245195dbf0816485f8230a16d02cf3aa39e8fed9dcffb8da85e87c8 + collected_at: '2026-09-17' + rows: 8856 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/gemm/vllm/0.25.0/gemm_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/gemm/vllm/0.25.0/gemm_perf.parquet new file mode 100644 index 0000000000..c58204b10e Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/gemm/vllm/0.25.0/gemm_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/linear_attention/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/linear_attention/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..f06ecf1968 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/linear_attention/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,21 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + gdn_perf: + rows: 10324 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:ad983a508da4af6b521f77f75043cd8959537105866666aa1fbdd8da1298212f + case_plan_hash: sha256:ba375a4c3a8dbbe8b77cec5cfd0f9dafd245e18b99ace763cc738c9645962dda + collected_at: '2026-09-17' + rows: 10324 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/linear_attention/vllm/0.25.0/gdn_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/linear_attention/vllm/0.25.0/gdn_perf.parquet new file mode 100644 index 0000000000..4326973883 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/linear_attention/vllm/0.25.0/gdn_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mhc/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mhc/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..5d7d5a32b8 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mhc/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,21 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + mhc_module_perf: + rows: 140 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:fc748ba7378714bee0e03107d3dec91f0d4659792df2cd2e31a51bdde4dd5c43 + case_plan_hash: sha256:4251dd697ea2e18fcaeed92b1a7cbdf452549a75d3a38284340d5da8502e03a1 + collected_at: '2026-09-17' + rows: 140 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mhc/vllm/0.25.0/mhc_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mhc/vllm/0.25.0/mhc_module_perf.parquet new file mode 100644 index 0000000000..e0fa765d09 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mhc/vllm/0.25.0/mhc_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mla/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mla/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..22811dd5ab --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mla/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,37 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + mla_context_module_perf: + rows: 5820 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:113fa413702792e5aa95bf4ee7c12f5da9b4410e97fe1be13054acc088a6b7fb + case_plan_hash: sha256:07e0ec5401fec176bb9e2ffe8d5ec19c54a2355f25ead16ff7e1960a279ca0df + collected_at: '2026-09-17' + rows: 5819 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:113fa413702792e5aa95bf4ee7c12f5da9b4410e97fe1be13054acc088a6b7fb + case_plan_hash: sha256:a6b2278aee53f798e8281c6078b2902e71ca03ff4ca9a4740e598e8e2956acd8 + collected_at: '2026-09-17' + rows: 1 + status: complete + mla_generation_module_perf: + rows: 8832 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:113fa413702792e5aa95bf4ee7c12f5da9b4410e97fe1be13054acc088a6b7fb + case_plan_hash: sha256:23f8a42b14c9eec372c32c809e900acb3fb4e5e1bad5c9adf14124dacd32943f + collected_at: '2026-09-17' + rows: 8832 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mla/vllm/0.25.0/mla_context_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mla/vllm/0.25.0/mla_context_module_perf.parquet new file mode 100644 index 0000000000..14bab13329 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mla/vllm/0.25.0/mla_context_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mla/vllm/0.25.0/mla_generation_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mla/vllm/0.25.0/mla_generation_module_perf.parquet new file mode 100644 index 0000000000..8c09a51f5e Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mla/vllm/0.25.0/mla_generation_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mla_bmm/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mla_bmm/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..04e02dce62 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mla_bmm/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,27 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + mla_bmm_perf: + rows: 636 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:4c2fd32c623c9841221e74bf76389706de4171beb0c22793246b0ce7ea79af87 + case_plan_hash: sha256:3928a5f969ae055ba17244c0d53b504e8253870ffcbf60696731071ddb98e1e5 + collected_at: '2026-09-17' + rows: 300 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:4c2fd32c623c9841221e74bf76389706de4171beb0c22793246b0ce7ea79af87 + case_plan_hash: sha256:faeaa20a84179b59df5283aa9a238572a431f4f16f8521470519147073fe4b53 + collected_at: '2026-09-17' + rows: 336 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mla_bmm/vllm/0.25.0/mla_bmm_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mla_bmm/vllm/0.25.0/mla_bmm_perf.parquet new file mode 100644 index 0000000000..b106076768 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/mla_bmm/vllm/0.25.0/mla_bmm_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/moe/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/moe/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..653b06d2f7 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/moe/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,27 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + moe_perf: + rows: 54432 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:fe2d2d23d340ded7bff27c17b5682e950458c2946d84d1a04bf8608cdb5a689d + case_plan_hash: sha256:86632275c959e534d7bc4d5ee15cd3b7cf60a75638a57992d14529f70f5e4b61 + collected_at: '2026-09-17' + rows: 54375 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:fe2d2d23d340ded7bff27c17b5682e950458c2946d84d1a04bf8608cdb5a689d + case_plan_hash: sha256:21f6decdc1e771534fcd759d0c1d2186a57c05fd5c1b1e10adf88112625e1915 + collected_at: '2026-09-17' + rows: 108 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/moe/vllm/0.25.0/moe_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/moe/vllm/0.25.0/moe_perf.parquet new file mode 100644 index 0000000000..2c151395a6 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/moe/vllm/0.25.0/moe_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/msa/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/msa/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..5bacf84547 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/msa/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,27 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + msa_context_module_perf: + rows: 11268 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:31c1c4310a284442a8275ef4c55c51eb54ff9ba934f18f3ceb9e2b5d002f2623 + case_plan_hash: sha256:2fa04db807438446fe3c239cafe10fc48f75b102de2146fa128bbbba094d9a90 + collected_at: '2026-09-17' + rows: 11199 + status: complete + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:31c1c4310a284442a8275ef4c55c51eb54ff9ba934f18f3ceb9e2b5d002f2623 + case_plan_hash: sha256:f7418cc760237ef166e719d63c1b083cebe1d0970b1782a6b1cd1486fba3e909 + collected_at: '2026-09-17' + rows: 69 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/msa/vllm/0.25.0/msa_context_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/msa/vllm/0.25.0/msa_context_module_perf.parquet new file mode 100644 index 0000000000..c6f1e331b1 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/msa/vllm/0.25.0/msa_context_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/collection_meta.yaml b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/collection_meta.yaml new file mode 100644 index 0000000000..0189e099c3 --- /dev/null +++ b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/collection_meta.yaml @@ -0,0 +1,71 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +schema_version: 2 +provenance: local +runtime: + framework: vllm + version: 0.25.0 + image: vllm/vllm-openai + image_digest: sha256:e1c1ff1af9a15921bfa11d1d95047258c1797392cdbfa296e7639da446b23f97 +tables: + dsv4_csa_context_module_perf: + rows: 5840 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:38a6cd429db806b370f13a9b847ad836440166e29d1e68c8e0c56da13396d5db + collected_at: '2026-09-17' + rows: 5840 + status: complete + dsv4_csa_generation_module_perf: + rows: 1544 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:21f9356813701f48d6451510032ec2746c9a64e3433c7f9954a0156c187dcae0 + collected_at: '2026-09-17' + rows: 1544 + status: complete + dsv4_hca_attn_module_perf: + rows: 3918 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:0e844f577823301a0af954d0d20baab4642bb1c0368652998bc28160dfab816d + collected_at: '2026-09-17' + rows: 3918 + status: complete + dsv4_hca_context_module_perf: + rows: 5840 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:d36420598033f22968df6d517a051324bdac1f448ff41be9aacf12beec3dd893 + collected_at: '2026-09-17' + rows: 5840 + status: complete + dsv4_hca_generation_module_perf: + rows: 1544 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:f6f9fe724d2fa5ad60fb6d0a162f93cd7901bd26093925a5fef0cbae9f310eb4 + collected_at: '2026-09-17' + rows: 1544 + status: complete + dsv4_paged_mqa_logits_module_perf: + rows: 3918 + status: complete + collections: + - collector_ref: cbaf51b64fa460e5ec6146bde407a4c64958212d + collector_hash: sha256:5882781b43c67c93f89e3e238978f461e242c83f681effd9ee6129a4be1637eb + case_plan_hash: sha256:0b35d1427f37f58a74ae69d6503f0eecb1ba5d60499325b10e1dad10a8199250 + collected_at: '2026-09-17' + rows: 3918 + status: complete diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_context_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_context_module_perf.parquet new file mode 100644 index 0000000000..f621179c8d Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_context_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_generation_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_generation_module_perf.parquet new file mode 100644 index 0000000000..2fa778a450 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_csa_generation_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_attn_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_attn_module_perf.parquet new file mode 100644 index 0000000000..d865bf5b4c Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_attn_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_context_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_context_module_perf.parquet new file mode 100644 index 0000000000..7e1a54c03b Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_context_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_generation_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_generation_module_perf.parquet new file mode 100644 index 0000000000..8b5963433f Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_hca_generation_module_perf.parquet differ diff --git a/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_paged_mqa_logits_module_perf.parquet b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_paged_mqa_logits_module_perf.parquet new file mode 100644 index 0000000000..cde56fab95 Binary files /dev/null and b/python/aisimulate/src/aiconfigurator_core/systems/data/h200_sxm/sparse_attention/vllm/0.25.0/dsv4_paged_mqa_logits_module_perf.parquet differ diff --git a/python/aisimulate/tests/unit/sdk/database/test_reuse_ordering.py b/python/aisimulate/tests/unit/sdk/database/test_reuse_ordering.py index 0e201e61ae..a1f43ccd69 100644 --- a/python/aisimulate/tests/unit/sdk/database/test_reuse_ordering.py +++ b/python/aisimulate/tests/unit/sdk/database/test_reuse_ordering.py @@ -27,12 +27,14 @@ import logging from pathlib import Path +import pyarrow.parquet as pq import pytest import yaml from aiconfigurator.sdk import common from aiconfigurator.sdk.operations.base import resolve_op_data_path -from aiconfigurator.sdk.perf_database import PerfDatabase +from aiconfigurator.sdk.perf_database import PerfDatabase, get_database +from aiconfigurator_core.sdk.engine_table_view import fetch_table_view pytestmark = pytest.mark.unit @@ -862,3 +864,28 @@ def test_vetoed_primary_with_no_donor_loads_nothing_through_the_engine_view(syst view = fetch_table_view(db, "_gemm_data") assert view is None or not view, f"the vetoed primary leaked into the engine view: {view!r}" + + +@pytest.mark.parametrize("system", ["b200_sxm", "b300_sxm", "gb200", "gb300", "h100_sxm", "h200_sxm"]) +def test_corrected_024_gemm_uses_declared_025_measurements(system): + """Every unshadowed donor row loads; retained 0.24 primary latencies win.""" + data = Path(__file__).resolve().parents[4] / "src/aiconfigurator_core/systems/data" / system + old = pq.read_table(data / "gemm/vllm/0.24.0/gemm_perf.parquet").to_pylist() + fresh = pq.read_table(data / "gemm/vllm/0.25.0/gemm_perf.parquet").to_pylist() + assert all(row["gemm_dtype"] != "fp8_block" for row in old) + db = get_database(system, "vllm", "0.24.0", shared_layer=True, strict_provenance=True) + loaded = fetch_table_view(db, "_gemm_data") + sources = db.data_provenance["gemm_perf.parquet"] + assert [(s["version"], s["channel"]) for s in sources[:2]] == [("0.24.0", "primary"), ("0.25.0", "declared_reuse")] + expected = {} + for row in old + fresh: + key = (common.GEMMQuantMode[row["gemm_dtype"]], row["m"], row["n"], row["k"]) + expected.setdefault(key, row["latency"]) + actual = { + (mode, m, n, k): value["latency"] + for mode, ms in loaded.items() + for m, ns in ms.items() + for n, ks in ns.items() + for k, value in ks.items() + } + assert actual == expected diff --git a/scripts/audit_pr244_data.py b/scripts/audit_pr244_data.py new file mode 100644 index 0000000000..a396868261 --- /dev/null +++ b/scripts/audit_pr244_data.py @@ -0,0 +1,210 @@ +#!/usr/bin/env python3 +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 +"""Reproduce PR #244's sanitized collection and parquet evidence without a GPU.""" + +import argparse +import gzip +import hashlib +import json +import re +import subprocess +from pathlib import Path + +import pyarrow as pa +import pyarrow.compute as pc +import pyarrow.parquet as pq + +ROOT = Path(__file__).resolve().parents[1] +DATA = ROOT / "python/aisimulate/src/aiconfigurator_core/systems/data" +EVIDENCE = ROOT / "docs/data/pr244" +SOURCE = "1f29ee459882796db312b8a288c040b83edbb211" +SYSTEMS = ("b300_sxm", "gb200", "gb300", "h100_sxm", "h200_sxm") + + +def sha256(data): + return hashlib.sha256(data).hexdigest() + + +def plan_hash(ids): + return sha256("\n".join(ids).encode()) + + +def require(condition, message): + if not condition: + raise ValueError(message) + + +def extract_cases(): + """Allowlist final checkpoint IDs/statuses; never export diagnostic text.""" + ledger = {"schema_version": 1, "source_revision": SOURCE, "plans": {}, "systems": {}} + for system in SYSTEMS: + prefix = f"{DATA.relative_to(ROOT)}/{system}/vllm-0.25.0-" + raw = { + name: subprocess.check_output(["git", "show", f"{SOURCE}:{prefix}{name}"], cwd=ROOT) + for name in ("collection-report.json", "failures.json.gz") + } + report = json.loads(raw["collection-report.json"]) + failures = json.loads(gzip.decompress(raw["failures.json.gz"])) + attempts = [] + for attempt in report["attempts"]: + group = failures[attempt["id"]] + done, failed = set(), set() + for checkpoint in group["checkpoints"].values(): + done.update(checkpoint["done"]) + failed.update(checkpoint["failed"]) + require(not done & failed, f"Conflicting final outcomes: {system}/{attempt['id']}") + # B300's archived plan was omitted. Its final checkpoint union must + # match BOTH the independently recorded retained count and ID digest. + ids = sorted(group.get("case_plan", {}).get("task_ids", done | failed)) + digest = plan_hash(ids) + require(digest == attempt["task_ids_sha256"], "Original retained-plan digest mismatch") + require(len(ids) == attempt["expected"] == len(set(ids)), "Original retained-plan count mismatch") + require(done | failed <= set(ids), "Outcome outside the retained plan") + for task in ids: + require( + re.fullmatch(r"(?:vllm\.[a-z0-9_]+:[a-z0-9_]+:\[[a-zA-Z0-9_/., '\-\[\]]*\]|mhc_[a-z0-9_]+)", task), + f"Unexpected task-ID format: {system}/{attempt['id']}", + ) + statuses = "".join("D" if task in done else "F" if task in failed else "U" for task in ids) + counts = { + "expected": len(ids), + "done": statuses.count("D"), + "failed": statuses.count("F"), + "unattempted": statuses.count("U"), + } + require(all(counts[key] == attempt[key] for key in counts), "Original outcome counts mismatch") + ledger["plans"][digest] = ids + attempts.append( + { + "id": attempt["id"], + "op": attempt["op"], + "plan_sha256": digest, + "plan_source": "archived_case_plan" if "case_plan" in group else "verified_checkpoint_union", + "reported_counts": counts, + "outcomes": statuses, + } + ) + ledger["systems"][system] = { + "source_artifact_sha256": {name: sha256(content) for name, content in raw.items()}, + "attempts": attempts, + "reported_tables": { + name: {key: table[key] for key in ("rows", "sha256")} for name, table in report["tables"].items() + }, + } + EVIDENCE.mkdir(parents=True, exist_ok=True) + payload = json.dumps(ledger, sort_keys=True, separators=(",", ":")).encode() + (EVIDENCE / "case-outcomes.json.gz").write_bytes(gzip.compress(payload, mtime=0)) + + +def summarize_cases(ledger): + for digest, ids in ledger["plans"].items(): + require(ids == sorted(set(ids)) and plan_hash(ids) == digest, "Retained-plan digest/order mismatch") + systems = {} + for system in SYSTEMS: + attempts = [] + seen = set() + for attempt in ledger["systems"][system]["attempts"]: + require(attempt["id"] not in seen, "Duplicate attempt ID") + seen.add(attempt["id"]) + ids = ledger["plans"][attempt["plan_sha256"]] + outcomes = attempt["outcomes"] + require(len(outcomes) == len(ids) and set(outcomes) <= set("DFU"), "Invalid outcome vector") + counts = { + "expected": len(ids), + "done": outcomes.count("D"), + "failed": outcomes.count("F"), + "unattempted": outcomes.count("U"), + } + require(counts == attempt["reported_counts"], "Reported outcome counts mismatch") + attempts.append({key: attempt[key] for key in ("id", "op", "plan_sha256", "plan_source")} | counts) + systems[system] = { + "cases": {key: sum(a[key] for a in attempts) for key in ("expected", "done", "failed", "unattempted")}, + "attempts": attempts, + } + return systems + + +def summarize_table(path): + table = pq.read_table(path) + # Physical row identity used by the original collection validation; this is + # not a claim about the engine's normalized lookup keys. + keys = [name for name in table.column_names if name not in ("device", "latency")] + columns = {} + for field in table.schema: + column = table[field.name] + summary = {"type": str(field.type), "nulls": column.null_count, "distinct": pc.count_distinct(column).as_py()} + if pa.types.is_integer(field.type) or pa.types.is_floating(field.type): + summary.update(pc.min_max(column).as_py()) + else: + summary["values"] = sorted(value for value in pc.unique(column).to_pylist() if value is not None) + columns[field.name] = summary + latency = table["latency"] + key_rows = zip(*(table[key].to_pylist() for key in keys), strict=True) + return { + "path": str(path.relative_to(ROOT)), + "sha256": sha256(path.read_bytes()), + "rows": table.num_rows, + "sidecar_sha256": sha256(path.with_name("collection_meta.yaml").read_bytes()), + "physical_key_columns": keys, + "latency_unit": "ms", + "columns": columns, + "anomalies": { + "duplicate_physical_keys": table.num_rows - len(set(key_rows)), + "null_cells": sum(column.null_count for column in table.columns), + "nonfinite_latency": pc.sum(pc.invert(pc.is_finite(latency))).as_py(), + "nonpositive_latency": pc.sum(pc.less_equal(latency, 0)).as_py(), + }, + } + + +def build_manifest(): + raw = (EVIDENCE / "case-outcomes.json.gz").read_bytes() + ledger = json.loads(gzip.decompress(raw)) + require(ledger["source_revision"] == SOURCE, "Unexpected source revision") + systems = summarize_cases(ledger) + for system, summary in systems.items(): + tables = {path.name: summarize_table(path) for path in sorted((DATA / system).glob("*/vllm/0.25.0/*.parquet"))} + reported = ledger["systems"][system]["reported_tables"] + require(set(tables) == set(reported), f"Table inventory drift: {system}") + for name, table in tables.items(): + require( + all(count == 0 for count in table["anomalies"].values()), + f"Published table contains anomalies: {system}/{name}: {table['anomalies']}", + ) + require( + all(table[key] == reported[name][key] for key in ("rows", "sha256")), + f"Published table differs from collection report: {system}/{name}", + ) + summary.update({"table_count": len(tables), "rows": sum(t["rows"] for t in tables.values()), "tables": tables}) + return { + "schema_version": 1, + "source_revision": SOURCE, + "collector_revision": "cbaf51b64fa460e5ec6146bde407a4c64958212d", + "vllm_runtime_revision": "dd10e03f95f94edbea1975c67ace3a35ec9a8a40", + "case_outcomes_sha256": sha256(raw), + "systems": systems, + } + + +def main(): + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument( + "--extract-cases", action="store_true", help="Recover sanitized evidence from pinned Git history" + ) + parser.add_argument("--write", action="store_true", help="Write the manifest instead of checking it") + args = parser.parse_args() + if args.extract_cases: + extract_cases() + manifest = build_manifest() + path = EVIDENCE / "manifest.json" + if args.write: + path.write_text(json.dumps(manifest, indent=2, sort_keys=True, allow_nan=False) + "\n") + else: + require(manifest == json.loads(path.read_text()), "Manifest drift; inspect changes before using --write") + for system, summary in manifest["systems"].items(): + print(f"{system}: {summary['table_count']} tables, {summary['rows']} rows, {summary['cases']}") + + +if __name__ == "__main__": + main() diff --git a/scripts/tests/test_pr244_data_audit.py b/scripts/tests/test_pr244_data_audit.py new file mode 100644 index 0000000000..d9be44ba69 --- /dev/null +++ b/scripts/tests/test_pr244_data_audit.py @@ -0,0 +1,118 @@ +# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 + +"""Reject corrupt coverage evidence and expose bad timing rows.""" + +import copy +import gzip +import json +import tempfile +import unittest +from pathlib import Path +from unittest.mock import patch + +import pyarrow as pa +import pyarrow.parquet as pq + +from scripts import audit_pr244_data as audit + + +class CollectionEvidenceTest(unittest.TestCase): + def setUp(self): + ids = ["case-a", "case-b", "case-c"] + digest = audit.plan_hash(ids) + attempt = { + "id": "gemm", + "op": "gemm", + "plan_sha256": digest, + "plan_source": "archived_case_plan", + "outcomes": "DFU", + "reported_counts": {"expected": 3, "done": 1, "failed": 1, "unattempted": 1}, + } + self.ledger = { + "plans": {digest: ids}, + "systems": {system: {"attempts": [copy.deepcopy(attempt)]} for system in audit.SYSTEMS}, + } + + def test_unattempted_cases_are_counted(self): + summary = audit.summarize_cases(self.ledger) + self.assertEqual(summary["b300_sxm"]["cases"], {"expected": 3, "done": 1, "failed": 1, "unattempted": 1}) + + def test_rejects_tampered_plan_and_outcomes(self): + mutations = [ + lambda ledger: next(iter(ledger["plans"].values())).append("case-d"), + lambda ledger: ledger["systems"]["b300_sxm"]["attempts"][0].update(outcomes="DF"), + lambda ledger: ledger["systems"]["b300_sxm"]["attempts"][0].update(outcomes="DDD"), + lambda ledger: ledger["systems"]["b300_sxm"]["attempts"][0].update(outcomes="DFX"), + lambda ledger: ledger["systems"]["b300_sxm"]["attempts"].append( + ledger["systems"]["b300_sxm"]["attempts"][0] + ), + ] + for mutate in mutations: + ledger = copy.deepcopy(self.ledger) + mutate(ledger) + with self.assertRaises(ValueError): + audit.summarize_cases(ledger) + + def test_reports_duplicate_null_and_invalid_latency_rows(self): + with tempfile.TemporaryDirectory() as directory: + root = Path(directory) + path = root / "test.parquet" + pq.write_table( + pa.table( + { + "m": [1, 1, 2, 3], + "gemm_dtype": ["fp8", "fp8", None, "fp8"], + "latency": [1.0, -1.0, float("inf"), 0.0], + } + ), + path, + ) + path.with_name("collection_meta.yaml").write_text("schema_version: 2\n") + with patch.object(audit, "ROOT", root): + result = audit.summarize_table(path) + self.assertEqual( + result["anomalies"], + {"duplicate_physical_keys": 1, "null_cells": 1, "nonfinite_latency": 1, "nonpositive_latency": 2}, + ) + self.assertEqual(result["latency_unit"], "ms") + self.assertEqual(result["columns"]["m"]["max"], 3) + + def test_write_rejects_anomalies_even_when_report_matches(self): + cases = { + "duplicate_physical_keys": {"m": [1, 1]}, + "null_cells": {"m": [1, None]}, + "negative_latency": {"latency": [1.0, -1.0]}, + "zero_latency": {"latency": [1.0, 0.0]}, + "infinite_latency": {"latency": [1.0, float("inf")]}, + "nan_latency": {"latency": [1.0, float("nan")]}, + } + for label, invalid_columns in cases.items(): + with self.subTest(label=label), tempfile.TemporaryDirectory() as directory: + root = Path(directory) + data = root / "data" + path = data / "b300_sxm/gemm/vllm/0.25.0/gemm_perf.parquet" + path.parent.mkdir(parents=True) + columns = {"m": [1, 2], "gemm_dtype": ["fp8", "fp8"], "latency": [1.0, 2.0]} + pq.write_table(pa.table(columns | invalid_columns), path) + path.with_name("collection_meta.yaml").write_text("schema_version: 2\n") + ledger = copy.deepcopy(self.ledger) + ledger["source_revision"] = audit.SOURCE + ledger["systems"]["b300_sxm"]["reported_tables"] = { + path.name: {"rows": 2, "sha256": audit.sha256(path.read_bytes())} + } + (root / "case-outcomes.json.gz").write_bytes(gzip.compress(json.dumps(ledger).encode())) + manifest = root / "manifest.json" + original = '{"previous": "valid manifest"}\n' + manifest.write_text(original) + with ( + patch.multiple(audit, ROOT=root, DATA=data, EVIDENCE=root, SYSTEMS=("b300_sxm",)), + patch("sys.argv", ["audit_pr244_data.py", "--write"]), + self.assertRaisesRegex(ValueError, "Published table contains anomalies: b300_sxm/gemm_perf"), + ): + audit.main() + self.assertEqual(manifest.read_text(), original) + + +if __name__ == "__main__": + unittest.main()