74 Commits

Author SHA1 Message Date
ElizaWszola
2e7035dd8c
[Bugfix] Fix fp8 DeepGemm compilation issues (#30336) 2025-12-09 20:17:25 -05:00
Charlie Fu
3c680f4a17
[Rocm][torch.compile] Adding layernorm + fp8 block quant and silu + fp8 block quant for Aiter (#25693)
Signed-off-by: charlifu <charlifu@amd.com>
Signed-off-by: Micah Williamson <micah.williamson@amd.com>
Signed-off-by: Charlie Fu <Charlie.Fu@amd.com>
Co-authored-by: Micah Williamson <micah.williamson@amd.com>
Co-authored-by: wuhuikx <hattie.wu@amd.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
Co-authored-by: Gregory Shtrasberg <156009573+gshtras@users.noreply.github.com>
2025-12-09 22:39:26 +00:00
Kyle Sayers
fccd532587
[Quantization] FP8 Weight Reloading for Quantized RL Rollout (#28480)
Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>
2025-12-09 13:54:32 -08:00
Zhewen Li
ae339b1a67
[Bugfix] Fix DeepGEMM after #29546 (#30267)
Signed-off-by: zhewenli <zhewenli@meta.com>
Signed-off-by: Zhewen Li <zhewenli@meta.com>
2025-12-09 01:05:27 +00:00
ElizaWszola
af0444bf40
[Performance] Fused blockwise quant RMS norm (#27883)
Signed-off-by: ElizaWszola <ewszola@redhat.com>
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Co-authored-by: yewentao256 <zhyanwentao@126.com>
2025-12-07 16:38:04 +00:00
Wentao Ye
541a2ef892
[Perf] Deepgemm fused layout kernel for activations, 4.3% throughput improvement, 10.7% TTFT improvement. (#29546)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
2025-12-07 20:31:14 +08:00
Varun Sundar Rabindranath
19bee6d12d
[Performance][DP/EP] Add silu_mul_per_token_group_quant_fp8_colmajor kernel (#29470)
Signed-off-by: Varun Sundar Rabindranath <vsundarr@redhat.com>
Co-authored-by: Varun Sundar Rabindranath <vsundarr@redhat.com>
Co-authored-by: Tyler Michael Smith <tyler@neuralmagic.com>
2025-12-03 18:04:59 +00:00
TJian
a425dc256e
[Bugfix] [ROCm] [AITER]: Fix aiter block quant not compatible with torch compile dynamo (#28716)
Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>
2025-11-14 10:30:50 -08:00
vllmellm
d8140b9833
[ROCM] Fix ROCm warnings, environment flag access, and GEMM kernel naming for consistency in _aiter_ops.py (#28464)
Signed-off-by: vllmellm <vllm.ellm@embeddedllm.com>
2025-11-12 21:46:57 +00:00
Varun Sundar Rabindranath
74a9a9faad
[Performance][B200] Fix deepgemm prologue (#27897)
Signed-off-by: Varun Sundar Rabindranath <vsundarr@redhat.com>
Co-authored-by: Varun Sundar Rabindranath <vsundarr@redhat.com>
2025-11-12 13:13:03 -08:00
Alexander Matveev
f76e85c299
[Performance][Hopper] Avoid M dim padding to 4x for most cases (due to cuda graphs paddings) (#28492)
Signed-off-by: Alexander Matveev <amatveev@redhat.com>
2025-11-12 10:51:43 -05:00
Michael Goin
f9a4087182
Remove weight_scale.T special case for SM90 Block FP8 CUTLASS kernel (#28431)
Signed-off-by: mgoin <mgoin64@gmail.com>
2025-11-11 11:46:04 -05:00
Yong Hoon Shin
021143561f
[ROCm] Add missing gemm_a8w8_blockscale import (#28378)
Signed-off-by: Yong Hoon Shin <yhshin@meta.com>
2025-11-10 23:13:36 +00:00
vllmellm
f080a83511
[RFC][ROCm][AITER] Keep all AITER kernels in _aiter_ops class like _custom_ops and _ipex_ops (#24490)
Signed-off-by: vllmellm <vllm.ellm@embeddedllm.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
2025-11-10 08:20:53 -08:00
Aleksandr Malyshev
449de9001a
[ROCm] triton fp8 kernel (#27058)
Signed-off-by: Aleksandr Malyshev <maleksan@amd.com>
Co-authored-by: Aleksandr Malyshev <maleksan@amd.com>
Co-authored-by: Gregory Shtrasberg <156009573+gshtras@users.noreply.github.com>
2025-11-06 14:46:44 -05:00
Isotr0py
6ac5e06f7c
[Chore] Clean up pytorch helper functions in vllm.utils (#26908)
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
Signed-off-by: isotr0py <2037008807@qq.com>
2025-10-18 09:48:22 -07:00
Harry Mellor
6c9fdbf725
[Docs] Replace rst style double-backtick with md single-backtick (#27091)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
2025-10-17 02:47:34 -07:00
Harry Mellor
8fcaaf6a16
Update Optional[x] -> x | None and Union[x, y] to x | y (#26633)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
2025-10-12 09:51:31 -07:00
Lucas Kabela
213b64452a
[Bugfix] Convert untraceable GroupShape to list for AMD impl (#26535)
Signed-off-by: Lucas Kabela <lucaskabela@meta.com>
2025-10-10 13:32:29 +00:00
Harry Mellor
d6953beb91
Convert formatting to use ruff instead of yapf + isort (#26247)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
2025-10-05 07:06:22 -07:00
ElizaWszola
502640c3f9
[Perf] Fix and reapply move apply w8a8 block fp8 linear to class (#25696)
Signed-off-by: ElizaWszola <ewszola@redhat.com>
Signed-off-by: ElizaWszola <elizaw.9289@gmail.com>
Signed-off-by: Luka Govedič <lgovedic@redhat.com>
Signed-off-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
Co-authored-by: Luka Govedič <lgovedic@redhat.com>
2025-10-02 19:35:13 +00:00
Wentao Ye
89e4050af4
[Bug] Fix Weight Loading for Block FP8 Cutlass SM90 (#25909)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Signed-off-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
2025-09-30 09:15:19 +08:00
Wentao Ye
9fe4c2bdb9
[Refactor] Remove DeepGEMM OP Register (#25710)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
2025-09-25 20:13:41 -04:00
Tyler Michael Smith
1260180c67
Revert "[Performance] Move apply_w8a8_block_fp8_linear to an op class… (#25607)
Signed-off-by: Tyler Michael Smith <tlrmchlsmth@gmail.com>
2025-09-25 08:05:21 +00:00
Wentao Ye
1f29141258
[Refactor] Use DeepGEMM Col Major TMA Aligned Tensor (#25517)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
2025-09-24 18:52:36 -04:00
Li, Jiang
1cbcfb94de
[Bugfix][CPU] Skip unsupported custom op register on CPU (#25534)
Signed-off-by: jiang1.li <jiang1.li@intel.com>
2025-09-24 06:21:51 +00:00
Michael Goin
7361ab379f
Remove redundant mutates_args and dispatch_key for direct_register_custom_op (#25512)
Signed-off-by: mgoin <mgoin64@gmail.com>
2025-09-23 22:48:40 +00:00
ElizaWszola
63400259d0
[Performance] Move apply_w8a8_block_fp8_linear to an op class (#24666)
Signed-off-by: ElizaWszola <ewszola@redhat.com>
Signed-off-by: ElizaWszola <elizaw.9289@gmail.com>
Signed-off-by: Luka Govedič <lgovedic@redhat.com>
Signed-off-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
Co-authored-by: Luka Govedič <lgovedic@redhat.com>
2025-09-23 12:03:10 -07:00
bnellnm
f11e3c516b
[Kernels] Support blocked fp8 quantization for compressed tensors MoE (#25219)
Signed-off-by: Bill Nell <bnell@redhat.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
2025-09-23 16:11:34 +00:00
Michael Goin
fbd6523ac0
Refactor dense FP8 tensor/channel/block utils and add CT FP8 block (#21404) 2025-09-18 08:53:45 -04:00
bnellnm
5963b98b46
[Kernel] Delegate construction of FusedMoEQuantConfig to FusedMoEMethodBase subclasses (#22537)
Signed-off-by: Bill Nell <bnell@redhat.com>
2025-09-17 17:43:31 -06:00
Michael Goin
c3aea10dc8
[Perf] Use upstream CUTLASS for SM90 Block FP8 kernel (#23280)
Signed-off-by: mgoin <mgoin64@gmail.com>
Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
2025-09-11 15:43:14 -07:00
Wentao Ye
3af47c3cc6
[Feature] Add Hopper DeepGEMM E8M0 for DeepSeekV3.1 scale_fmt (#23666)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Signed-off-by: youkaichao <youkaichao@gmail.com>
Co-authored-by: youkaichao <youkaichao@gmail.com>
2025-08-27 14:09:08 +00:00
Wentao Ye
394591e343
[Feature] Enable DeepGEMM Linear on B200; 1.5% E2E throughput improvement (#23351)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
2025-08-21 21:01:08 -07:00
Wentao Ye
f7dcce7a4a
[Feature] Add VLLM_USE_DEEP_GEMM_E8M0 Env to Control E8M0 Scale (#21968)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
2025-08-11 09:39:08 -07:00
Wentao Ye
eec890c1c1
[Bug] Fix B200 DeepGEMM E8M0 Accuracy Issue (#22399)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
2025-08-06 17:03:53 -07:00
TJian
e626d286f5
[FEAT] [ROCm] [AITER]: Add AITER HIP block quant kernel (#21242) 2025-07-28 05:07:06 +00:00
Wentao Ye
633f6e804b
[Bug] Fix DeepGemm Init Error (#21554)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
2025-07-24 20:07:22 -07:00
Wentao Ye
774d0c014b
[Perf] Cuda Kernel for Per Token Group Quant (#21083)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
2025-07-22 07:27:15 -07:00
Wentao Ye
76ddeff293
[Doc] Remove duplicate docstring (#21012)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
2025-07-15 20:09:13 -07:00
TJian
80d38b8ac8
[V1] [ROCm] [AITER] Upgrade AITER to commit 916bf3c and bugfix APIs (#20880)
Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>
2025-07-13 15:19:32 +00:00
Wentao Ye
42d440c22b
[Perf] Use Triton instead of Torch for DeepGEMM Per Token Group Quant (#20841)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
2025-07-12 19:38:45 -07:00
Wentao Ye
e2de455c34
[Feature] Integrate SM100 DeepGEMM support (#20087) 2025-07-10 20:18:05 -07:00
Li, Jiang
0ec3779df7
[Bugfix][CI/CD][CPU] Fix CPU CI tests (#20383)
Signed-off-by: jiang1.li <jiang1.li@intel.com>
2025-07-02 20:11:36 -07:00
Wentao Ye
4d36693687
[Refactor] Create a function util and cache the results for has_deepgemm, has_deepep, has_pplx (#20187)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
2025-06-28 22:06:38 +00:00
Wentao Ye
879f69bed3
[Refactor] Remove duplicate ceil_div (#20023)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
2025-06-25 05:19:09 +00:00
Wentao Ye
a6c4b87fbc
Revert "[Feature] Integrate new deepgemm (#19820)" (#20049)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
2025-06-24 19:45:22 -07:00
Wentao Ye
c6e3bba8e6
[Feature] Integrate new deepgemm (#19820)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
2025-06-24 12:51:56 -07:00
Varun Sundar Rabindranath
e5d35d62f5
[BugFix] Force registration of w8a8_block_fp8_matmul_deepgemm via lazy import (#19514)
Signed-off-by: Varun Sundar Rabindranath <vsundarr@redhat.com>
Co-authored-by: Varun Sundar Rabindranath <vsundarr@redhat.com>
2025-06-12 04:28:12 +00:00
artetaout
b8e809a057
[Kernel] Support deep_gemm for linear methods (#19085)
Signed-off-by: artetaout <lulala341@gmail.com>
2025-06-11 15:14:45 +08:00