Shengqi Chen and GitHub
a02155c787
Merge branch 'main' into cuda-arch-fixup
2026-07-21 09:33:48 +08:00
qli88 and GitHub
5de1add806
[feature]Add int4 quantization support for emulation moe backend ( #48451 )
...
Signed-off-by: Qiang Li <qiang.li2@amd.com >
2026-07-16 02:00:12 +00:00
1d99f0f421
[ROCm][BugFix] Triton W4A16 handling for GPTQ/AutoGPTQ qzeros layout ( #47770 )
...
Signed-off-by: giuseppegrossi <ggrossi@amd.com >
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
2026-07-15 11:55:47 -05:00
382bbd5144
[ROCm][Kernel] Add HybridW4A16LinearKernel: Triton prefill + HIP skinny decode ( #40977 )
...
Signed-off-by: Matthias Gehre <matthias.gehre@amd.com >
Co-authored-by: Andreas Karatzas <akaratza@amd.com >
2026-07-13 20:22:00 -07:00
Shengqi Chen and GitHub
83a7669827
Merge branch 'main' into cuda-arch-fixup
...
Signed-off-by: Shengqi Chen <harry-chen@outlook.com >
2026-07-07 00:13:53 +08:00
Spandan Tiwari and GitHub
b71218107f
[ROCm][Test] Fix test_per_token_group_quant_fp8 tolerance for 1-ULP FP8 rounding on gfx950 ( #46944 )
...
Signed-off-by: Spandan Tiwari <sptiwari@amd.com >
2026-07-05 18:02:30 -05:00
Shengqi Chen
a10e369f06
[Build] Address CUDA arch review comments
...
Fix CUDA arch warning tests so they do not depend on importing the stable libtorch extension, which keeps the warning coverage active in lightweight CI environments and avoids mypy treating a fixture value as a base class.
Use regex for the compiled-arch parser, apply formatter output, and make the CUTLASS grouped GEMM Python support query fall back to false when the op is unavailable or unimplemented in the current build.
Signed-off-by: Shengqi Chen <harry-chen@outlook.com >
2026-06-30 20:03:28 +08:00
Shengqi Chen and Codex
0f471a3088
[Build] Scope stable CUDA kernel feature macros
...
Keep optional stable CUDA kernel feature macros on the source files that consume them instead of adding them to VLLM_GPU_FLAGS. This avoids perturbing unrelated compile commands and invalidating more cache entries when optional kernel families change.
Also align CUTLASS grouped MoE support with the SM10x/SM11x family so Thor works under both CUDA 12 SM101 and CUDA 13 SM110 reporting, and remove the stale ENABLE_CUTLASS_MLA definition left after the old CUTLASS MLA path was deleted.
Co-authored-by: Codex <codex@openai.com >
Signed-off-by: Shengqi Chen <harry-chen@outlook.com >
2026-06-30 19:27:42 +08:00
Matt and GitHub
95e6442a6b
[Hardware][AMD][CI] Fix Kernels Quantization test timeout ( #46859 )
...
Signed-off-by: Matthew Wong <Matthew.Wong2@amd.com >
2026-06-26 15:19:16 -05:00
fxmarty-amd and GitHub
552a9dbe59
[NVFP4][Emulation] Fuse NVFP4 weight dequantization with compute in triton kernel for w13/w2 MOE MLP linears ( #44667 )
...
Signed-off-by: Felix Marty <Felix.Marty@amd.com >
2026-06-25 19:33:00 -07:00
b3a688cb9e
[ROCm] Fix OOB During Model Warmup With ROCM_ATTN and MRV2 ( #46548 )
...
Signed-off-by: Micah Williamson <micah.williamson@amd.com >
Signed-off-by: Matthew Wong <Matthew.Wong2@amd.com >
Co-authored-by: Matthew Wong <Matthew.Wong2@amd.com >
2026-06-24 10:53:21 -05:00
Andreas Karatzas and GitHub
549c7074cd
[ROCm][CI] Skip the MoE Marlin tile-padding helper assertion ( #46580 )
...
Signed-off-by: Andreas Karatzas <akaratza@amd.com >
2026-06-24 07:31:33 +00:00
37a682d392
[Kernel] Extend Marlin thread-tile padding to MoE (WNA16 + FP8/MXFP8) ( #45703 )
...
Signed-off-by: mgoin <mgoin64@gmail.com >
Co-authored-by: Claude <noreply@anthropic.com >
2026-06-23 11:45:10 -07:00
Mohammad Miadh Angkad and GitHub
d1a38c2762
[Kernel][Performance] Add FlashInfer cutedsl NVFP4 GEMM backend ( #42235 )
...
Signed-off-by: Mohammad Miadh Angkad <176301910+mmangkad@users.noreply.github.com >
2026-06-22 16:17:18 -04:00
JasonLi314 and GitHub
93bad11912
[Bugfix] Fix gridDim.y overflow for large row counts ( #45255 )
...
Signed-off-by: Jason Li <li.jason.cs@gmail.com >
2026-06-19 23:27:45 -04:00
Micah Williamson and GitHub
4a083cc858
[ROCm][CI] Pin test_rocm_compressed_tensors_w8a8 to TRITON_ATTN ( #46180 )
...
Signed-off-by: Micah Williamson <micah.williamson@amd.com >
2026-06-19 15:20:06 -05:00
Wentao Ye and GitHub
2659f60a1a
[Refactor] Remove dead quantization code and tests ( #45454 )
...
Signed-off-by: yewentao256 <zhyanwentao@126.com >
2026-06-17 16:12:01 -04:00
71b961dd35
[Perf] SM90 cutlass fp8 mm supports odd M by swap_ab, 180~290% kernel performance improvement ( #44572 )
...
Signed-off-by: yewentao256 <zhyanwentao@126.com >
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
2026-06-13 12:05:45 -07:00
Michael Goin and GitHub
9eaacb23ec
[Kernel] Consolidate Marlin thread-tile padding across all dense Marlin paths ( #45295 )
...
Signed-off-by: mgoin <mgoin64@gmail.com >
2026-06-12 13:46:21 -07:00
Isotr0py and GitHub
6635279d8a
[Migration] Migrate GGUF quantization support to plugin ( #39612 )
...
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn >
2026-06-12 12:02:21 -07:00
32daf56b42
[Refactor] Rename rocm_moe.py to rocm_moe_rdna.py ( #45011 )
...
Signed-off-by: JartX <sagformas@epdcenter.es >
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
2026-06-10 17:02:09 +08:00
062b05ff3a
[ROCm][Perf] Fused MoE W4A16 HIP kernel for AMD RDNA3 (gfx1100) ( #44075 )
...
Signed-off-by: JartX <sagformas@epdcenter.es >
Co-authored-by: TJian <tunjian.tan@embeddedllm.com >
2026-06-06 15:30:39 -05:00
Agata Dobrzyniewicz and GitHub
a947f7a420
[Kernel][Test] Extend lightning_attn and awq_triton kernel tests to XPU ( #43307 )
...
Signed-off-by: Dobrzyniewicz, Agata <agata.dobrzyniewicz@intel.com >
2026-06-04 14:25:59 -04:00
Michael Goin and GitHub
06ee2d8433
[Quant] Support compressed-tensors WNA8O8Int linears and WNInt embeddings ( #44340 )
...
Signed-off-by: mgoin <mgoin64@gmail.com >
2026-06-04 07:40:33 -07:00
Charlie Fu and GitHub
71df063c49
Enable perf_token_group_quant/_C_stable_libtorch for ROCm ( #42758 )
...
Signed-off-by: charlifu <charlifu@amd.com >
2026-06-02 23:23:28 -07:00
JartX and GitHub
0cff0741ff
[Kernel][ROCm] Native W4A16 kernel for AMD RDNA3 (gfx1100) — fp16 + bf16 ( #41394 )
...
Signed-off-by: JartX <sagformas@epdcenter.es >
2026-05-29 11:04:40 +00:00
68e07d5916
[Bug] Fix ci issue assert output_size is not None AssertionError ( #43261 )
...
Signed-off-by: yewentao256 <zhyanwentao@126.com >
Signed-off-by: Isotr0py <Isotr0py@outlook.com >
Co-authored-by: Isotr0py <Isotr0py@outlook.com >
2026-05-21 16:58:09 +08:00
Michael Goin and GitHub
5774aad9c5
[Perf][gpt-oss] Downgrade triton_kernels to v3.5.1 ( #43135 )
...
Signed-off-by: mgoin <mgoin64@gmail.com >
2026-05-20 14:13:12 -07:00
363fc84407
Integrate flashinfer b12x MoE and FP4 GEMM kernels for SM120/121 ( #40082 )
...
Signed-off-by: Meenakshi Venkataraman <meenakshiv@nvidia.com >
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com >
2026-05-20 17:21:11 +00:00
Wentao Ye and GitHub
37ece593c1
[Perf] Padded nvfp4 quant kernel to remove additional copy, 2.4%~5.7% e2e performance improvement ( #42774 )
...
Signed-off-by: yewentao256 <zhyanwentao@126.com >
2026-05-18 16:38:12 -07:00
06d020bb6e
[Bugfix] Fix SM121 (DGX Spark) exclusion from Marlin/CUTLASS FP8 paths ( #35568 )
...
Signed-off-by: Blake Ledden <blake@secondnaturecomputing.com >
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com >
Co-authored-by: Pavani Majety <pmajety@nvidia.com >
2026-05-15 10:59:00 -07:00
John Calderon and GitHub
8a2fc80b84
[CUDA][CUTLASS] Enable cutlass scaled mm for non-compatible sizes ( #41868 )
...
Signed-off-by: John Calderon <jcalderon@nvidia.com >
2026-05-08 15:58:05 -07:00
Yuwen Zhou and GitHub
809b98e5b7
[CPU] Add FP8 W8A16 linear support ( #41186 )
...
Signed-off-by: yuwenzho <yuwen.zhou@intel.com >
2026-05-06 07:05:27 +00:00
a3c83ff2fd
Faster per-token fp8 group quant packed kernel for blackwell ( #41326 )
...
Signed-off-by: Yongye Zhu <zyy1102000@gmail.com >
Co-authored-by: Roger Wang <hey@rogerw.io >
2026-04-30 18:09:55 -07:00
2917d6363a
[NVFP4][Hopper/AMD Instinct] Add Triton kernels for NVFP4 dequantization and QDQ emulation ( #40033 )
...
Signed-off-by: Felix Marty <Felix.Marty@amd.com >
Co-authored-by: Claude <noreply@anthropic.com >
2026-04-30 17:35:48 -04:00
sychen52 and GitHub
6b2b7bd0eb
Add nvfp4 support to reshape_and_cache_flash ( #37332 )
...
Signed-off-by: Shiyang Chen <shiychen@nvidia.com >
2026-04-17 07:28:00 -07:00
1a9353bb02
[MoE] Move GPT OSS Triton kernel experts into fused_moe/experts/ ( #39007 )
...
Signed-off-by: Robert Shaw <robertgshaw2@gmail.com >
Signed-off-by: Jackmin801 <ongjackm@gmail.com >
Co-authored-by: Robert Shaw <robertgshaw2@gmail.com >
2026-04-14 19:27:39 +00:00
Micah Williamson and GitHub
4f4f8eaa78
[ROCm][CI] Fix condition for test_per_token_group_quant_fp8_packed ( #39730 )
...
Signed-off-by: Micah Williamson <micah.williamson@amd.com >
2026-04-14 16:14:31 +00:00
Wei Zhao and GitHub
59b2f7b640
[Perf] Fuse Zero Initializer for FP8 DeepGemm Block Quant Kernel ( #39547 )
...
Signed-off-by: wzhao18 <wzhao18.sz@gmail.com >
2026-04-11 07:16:51 -07:00
d468322dc1
[Kernel][Hardware][AMD] Add TritonW4A16LinearKernel for ROCm ( #37352 )
...
Signed-off-by: jatseng-ai <jatseng@amd.com >
Signed-off-by: jatseng-ai <janet.tseng@amd.com >
Signed-off-by: Matthias Gehre <matthias.gehre@amd.com >
Co-authored-by: Claude <noreply@anthropic.com >
Co-authored-by: Matthias Gehre <matthias.gehre@amd.com >
2026-04-10 10:25:27 +00:00
Maral and GitHub
2e9034c998
[W8A8 Block Linear Refactor][2/N] Remove W8A8Fp8BlockLinearOp and adopt Fp8 block linear kernel selections. ( #33892 )
...
Signed-off-by: maral <maralbahari.98@gmail.com >
Signed-off-by: Maral <maralbahari.98@gmail.com >
2026-04-09 08:50:39 +08:00
Andreas Karatzas and GitHub
2df2c85be4
[Kernels][MoE] Fix legacy_routing to use bitmatrix-based routing path ( #38504 )
...
Signed-off-by: Andreas Karatzas <akaratza@amd.com >
2026-04-07 10:57:09 +08:00
zhangyiming and GitHub
1ac6694297
[OOT] Add OOT support for linear kernel. ( #37989 )
...
Signed-off-by: menogrey <1299267905@qq.com >
2026-03-31 14:33:21 +08:00
Kyle Sayers and GitHub
38364a7e32
[Sparse24] [Deprecation] Remove Sparse24 CT integration and kernels ( #36799 )
...
Signed-off-by: Kyle Sayers <kylesayrs@gmail.com >
2026-03-23 16:03:29 -04:00
87bd91892f
[MoE Refactor] Mxfp4 oracle rebased ( #37128 )
...
Signed-off-by: Yongye Zhu <zyy1102000@gmail.com >
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com >
2026-03-21 03:37:04 +00:00
1779c09898
[ROCm] Enable wvSplitK skinny GEMM kernel for RDNA4/gfx1x decode ( #34709 )
...
Signed-off-by: L.B.R. <lbr@mmonad.com >
Co-authored-by: L.B.R. <lbr@mmonad.com >
2026-03-20 10:11:23 -05:00
Terry Gao and GitHub
3e6a1e1686
[Custom Ops] Add functional + out variant for scaled_fp4_quant ( #34389 )
...
Signed-off-by: tianrengao <terrygao87@gmail.com >
2026-03-16 18:51:46 -04:00
Kunshang Ji and GitHub
53ec16a705
[Hardware] Replace torch.cuda.device_count/current_device/set_device API ( #36145 )
...
Signed-off-by: Kunshang Ji <jikunshang95@gmail.com >
Signed-off-by: Kunshang Ji <kunshang.ji@intel.com >
2026-03-12 07:57:47 -07:00
Hashem Hashemi and GitHub
721ae79f50
Improvements to wvSplitKrc skinny GEMM solution ( #34304 )
...
Signed-off-by: Hashem Hashemi <hashem.hashemi@amd.com >
2026-03-10 09:14:27 -07:00
Kunshang Ji and GitHub
66a2209645
[Hardware] Replace torch.cuda.synchronize() api with torch.accelerator.synchronize ( #36085 )
...
Signed-off-by: Kunshang Ji <kunshang.ji@intel.com >
2026-03-05 10:36:39 +00:00
