|
268b1c55ad
|
[MoE Refactor][13/N] Convert FI to Use PFNoEP (#31533)
Signed-off-by: Robert Shaw <robshaw@redhat.com>
Signed-off-by: Robert Shaw <robertgshaw2@gmail.com>
Signed-off-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
Co-authored-by: Robert Shaw <robshaw@redhat.com>
Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
|
2026-01-03 12:26:36 -08:00 |
|
|
Yongye ZhuandRobert Shaw
|
a3f2f40947
|
[MoE Refactor] Explicit construct mk for flashinfer bf16 kernel (#31504)
Signed-off-by: Yongye Zhu <zyy1102000@gmail.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
|
2026-01-02 13:54:50 -08:00 |
|
|
Yongye ZhuandRobert Shaw
|
5a468ff7c7
|
[MoE Refactor] Split invoke_fused_moe_kernel (#31050)
Signed-off-by: Yongye Zhu <zyy1102000@gmail.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
|
2026-01-02 13:47:15 -08:00 |
|
|
Andreas Karatzas
|
6ef770df7c
|
[MoE] Fix output_shape calculation in Attention layer to handle 3D query inputs (#31596)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
|
2026-01-02 15:46:23 +00:00 |
|
|
Kyuyeun Kim
|
cc410e8644
|
[Bugfix] Fix weight_loader v1 block scale (#31103)
Signed-off-by: Kyuyeun Kim <kyuyeunk@google.com>
|
2026-01-02 13:14:10 +08:00 |
|
|
Vaibhav Sourirajan
|
1f43c121d5
|
Remove unused use_marlin variable in Mxfp4MoEMethod (#31549)
Signed-off-by: vaibhav sourirajan <vs2787@columbia.edu>
|
2026-01-01 21:13:36 -08:00 |
|
|
Tmn07
|
ca179d0f64
|
[Bugfix] Fix activation quantization for compressed-tensors W4A16 (#31572)
Signed-off-by: Tmn07 <tmn0796@gmail.com>
|
2026-01-01 21:13:22 -08:00 |
|
|
Kevin McKay
|
5fff44064b
|
[Bugfix] Replace BaseException with specific exceptions in FLA utils (#31590)
Signed-off-by: c0de128 <kevin.mckay@outlook.com>
|
2026-01-01 19:27:54 -08:00 |
|
|
Qiu
|
84f6cd741b
|
[Mics] add pcp basic support to MoE model (#31003)
|
2025-12-30 20:01:29 -08:00 |
|
|
JartX
|
07728bf5cd
|
[BugFix] add select_gemm_impl on CompressedTensorsWNA16MoEMethod to support LoRA (#31453)
Signed-off-by: JartX <sagformas@epdcenter.es>
|
2025-12-30 11:20:15 -08:00 |
|
|
|
56f516254c
|
[Bugfix][ROCm] Fix Static Quant Issue (#31502)
Signed-off-by: Robert Shaw <robshaw@redhat.com>
Co-authored-by: Robert Shaw <robshaw@redhat.com>
Co-authored-by: TJian <tunjian.tan@embeddedllm.com>
|
2025-12-29 13:27:55 -08:00 |
|
|
Robert ShawandRobert Shaw
|
9152a30d8f
|
[MoE Refactor][12/N] Marlin Fp8 MoE Pure Function (#31499)
Signed-off-by: Robert Shaw <robshaw@redhat.com>
Co-authored-by: Robert Shaw <robshaw@redhat.com>
|
2025-12-29 13:27:00 -08:00 |
|
|
Roger Youngandxuebi
|
5bc664110f
|
Optimize QKNorm for MiniMax-M2/M2.1 (#31493)
Signed-off-by: xuebi <xuebi@minimaxi.com>
Co-authored-by: xuebi <xuebi@minimaxi.com>
|
2025-12-29 16:30:18 +00:00 |
|
|
Mamy Ratsimbazafy
|
b9793e6a8c
|
Add Fused MoE Triton kernels for GLM-4.5-Air, GLM-4.5v, GLM-4.6v on 2x RTX Pro 6000 (#31407)
Signed-off-by: Mamy Ratsimbazafy <mamy_github@numforge.co>
|
2025-12-28 08:38:33 -08:00 |
|
|
Jzz1943
|
0b6b701050
|
[Model] Add tuned triton fused_moe configs for Qwen3Moe on B200 (#31448)
Signed-off-by: Zhongze Jiang <jiangzhongze.jzz@ant-intl.com>
|
2025-12-28 08:38:07 -08:00 |
|
|
Robert ShawandRobert Shaw
|
727c41f3fd
|
[MoE Refactor][10/N] Cleanup Fp8 Process Weights After Loading (#31169)
Signed-off-by: Robert Shaw <robshaw@redhat.com>
Co-authored-by: Robert Shaw <robshaw@redhat.com>
|
2025-12-27 20:22:48 +00:00 |
|
|
Xinyu Chen
|
87f1b8ca2c
|
CustomOp: Unify aiter impl into GroupedTopk (#31221)
Signed-off-by: Xinyu Chen <xinyu1.chen@intel.com>
|
2025-12-26 12:44:29 -05:00 |
|
|
|
506eb0f454
|
[Bugfix] Remove dead block_quant_to_tensor_quant function (#31294)
Co-authored-by: yurekami <yurekami@users.noreply.github.com>
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
|
2025-12-24 17:22:48 +00:00 |
|
|
Kevin McKay
|
66c9887440
|
[Bugfix][Hardware][AMD] Fix FP8 dtype in silu_mul quantization (#31179)
Signed-off-by: c0de128 <kevin.mckay@outlook.com>
|
2025-12-24 10:37:11 -05:00 |
|
|
Asaf Joseph Gardin
|
34916ae37f
|
[Mamba] - Consolidate Mambas Attention Logic (#28133)
|
2025-12-23 21:57:00 +01:00 |
|
|
Pavani Majety
|
3e10262356
|
Revert "[SM100] Enable fp8 compute for prefill MLA (#30746)" (#31197)
Signed-off-by: Pavani Majety <pmajety@nvidia.com>
|
2025-12-22 18:15:33 -08:00 |
|
|
Robert ShawandRobert Shaw
|
b57b967386
|
[MoE Refactor][7/N] AITER MK (#31102)
Signed-off-by: Robert Shaw <robshaw@redhat.com>
Co-authored-by: Robert Shaw <robshaw@redhat.com>
|
2025-12-22 16:42:58 -07:00 |
|
|
Pavani Majety
|
b10f41c894
|
[SM100] Enable fp8 compute for prefill MLA (#30746)
Signed-off-by: Pavani Majety <pmajety@nvidia.com>
|
2025-12-22 19:15:57 +00:00 |
|
|
Yongye Zhu
|
7b926e8901
|
[MoE Refactor][9/N] Use modular kernel for unquantized Triton MoE (#31052)
Signed-off-by: Yongye Zhu <zyy1102000@gmail.com>
|
2025-12-22 17:34:19 +00:00 |
|
|
Kevin McKayandClaude Sonnet 4.5
|
cf8eed7bef
|
[Bugfix][ROCm] Fix typo: is_linear_fp8_enaled -> is_linear_fp8_enabled (#31109)
Signed-off-by: c0de128 <kevin.mckay@outlook.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
|
2025-12-21 21:14:58 -08:00 |
|
|
CedricHuang
|
19cc9468fd
|
[Feature]: Support NVIDIA ModelOpt HF FP8 variants FP8_PER_CHANNEL_PER_TOKEN and FP8_PB_WO in vLLM (#30957)
|
2025-12-21 22:34:49 -05:00 |
|
|
Robert Shaw
|
b471092d3a
|
[MoE Refactor][4/N] Marlin Fp8 Mk (#31036)
|
2025-12-21 12:37:42 -05:00 |
|
|
Jinzhen Lin
|
7c73ceb581
|
[Quantization] add marlin w4a8/w8a8 check (#31061)
Signed-off-by: Jinzhen Lin <jinzhen.ljz@antgroup.com>
|
2025-12-20 21:58:11 +00:00 |
|
|
Jinzhen Linandgemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
|
ee52d9901d
|
[Quantization] support logical_widths for fp8 marlin (#30962)
Signed-off-by: Jinzhen Lin <jinzhen.ljz@antgroup.com>
Signed-off-by: Jinzhen Lin <linjinzhen@hotmail.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
|
2025-12-20 12:02:57 -08:00 |
|
|
|
54c8924384
|
[MoE Refactor][5/N] Isolate zero expert to LongCatFlash (#28891)
Signed-off-by: baonudesifeizhai <85092850+baonudesifeizhai@users.noreply.github.com>
Signed-off-by: Dongjie Zou <85092850+baonudesifeizhai@users.noreply.github.com>
Signed-off-by: baonudesifeizhai <baonudesifeizhai@gmail.com>
Signed-off-by: Robert Shaw <robertgshaw2@gmail.com>
Co-authored-by: Robert Shaw <robshaw@redhat.com>
Co-authored-by: Robert Shaw <robertgshaw2@gmail.com>
|
2025-12-20 18:22:04 +00:00 |
|
|
Yan Ma
|
560ae9638c
|
[XPU] enable fp8 online streaming quantization (#30944)
Signed-off-by: Yan Ma <yan.ma@intel.com>
|
2025-12-20 13:45:27 +00:00 |
|
|
Robert ShawandRobert Shaw
|
95befecc18
|
[MoE Refactor][2/N] Use Modular Kernels for Fp8 (#30825)
Signed-off-by: Robert Shaw <robshaw@redhat.com>
Co-authored-by: Robert Shaw <robshaw@redhat.com>
|
2025-12-19 23:36:38 +00:00 |
|
|
Robert ShawandRobert Shaw
|
83a317f650
|
[MoE Refactor][3/N] Deprecate cutlass block quant fp8 (b200) (#30990)
Signed-off-by: Robert Shaw <robshaw@redhat.com>
Co-authored-by: Robert Shaw <robshaw@redhat.com>
|
2025-12-19 13:09:54 -08:00 |
|
|
Wentao Ye
|
3bd8335bd0
|
[Refactor] Refactor for DeepGemmQuantScaleFMT using cache (#30898)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
|
2025-12-19 13:50:39 -07:00 |
|
|
|
969bbc7c61
|
[Model] Add MiMo-V2-Flash support (#30836)
Signed-off-by: Abatom <abzhonghua@gmail.com>
Signed-off-by: Jumiar <liuanqim10@126.com>
Signed-off-by: Zyann7 <zyann7@outlook.com>
Co-authored-by: Jumiar <liuanqim10@126.com>
Co-authored-by: Zyann7 <zyann7@outlook.com>
Co-authored-by: Jee Jee Li <pandaleefree@gmail.com>
|
2025-12-19 17:17:03 +00:00 |
|
|
Jinzhen Lin
|
5fbfa8d9ef
|
[Quantization] fix marlin w8a8 check (#30961)
Signed-off-by: Jinzhen Lin <jinzhen.ljz@antgroup.com>
|
2025-12-19 07:33:22 -08:00 |
|
|
Shanshan Shen
|
23a1946e3b
|
[CustomOp][Refactor] Extract common methods for ApplyRotaryEmb CustomOp (#31021)
Signed-off-by: shen-shanshan <467638484@qq.com>
|
2025-12-19 22:16:09 +08:00 |
|
|
Jinzhen Lin
|
9187de9fac
|
[Quantization] enable compressed-tensors marlin support for turing (2) (#31008)
Signed-off-by: Jinzhen Lin <jinzhen.ljz@antgroup.com>
|
2025-12-19 08:56:35 +00:00 |
|
|
Jinzhen Lin
|
de08b8f61b
|
[Quantization] enable compressed-tensors marlin support for turing (#31000)
Signed-off-by: Jinzhen Lin <jinzhen.ljz@antgroup.com>
|
2025-12-18 20:29:48 -08:00 |
|
|
Wentao Ye
|
97000a2be7
|
[Bug] Fix compressed tensor not using deepgemm (#30820)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
|
2025-12-18 14:45:55 -05:00 |
|
|
navmarri14
|
b8c477c115
|
tuned fused configs for B300 (#30629)
|
2025-12-18 11:41:59 -08:00 |
|
|
jiahanc
|
53ad423f26
|
[Perf] enable flashinfer rotary_embedding custom ops in DeepSeek rotary (#30729)
Signed-off-by: jiahanc <173873397+jiahanc@users.noreply.github.com>
|
2025-12-18 14:31:18 -05:00 |
|
|
Vasiliy Kuznetsov
|
f4ee2c3d90
|
fix fp8 online quantization streaming with tp > 1 (#30900)
Signed-off-by: vasiliy <vasiliy@fb.com>
|
2025-12-18 11:45:15 -05:00 |
|
|
Xin YangandCyrus Leung
|
9a5e96523b
|
[LoRA] Set default MXFP4 LoRA backend to Marlin (#30598)
Signed-off-by: Xin Yang <xyangx@amazon.com>
Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk>
|
2025-12-18 08:42:22 -08:00 |
|
|
Wentao YeandCyrus Leung
|
6628758233
|
[Bug] Fix batch invariant in torch 2.10 (#30907)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Signed-off-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk>
|
2025-12-18 07:27:51 -08:00 |
|
|
Ming Yang
|
8372be2828
|
[moe] Use enable_chunking func (to support disabling chunking) (#29935)
Signed-off-by: Ming Yang <minos.future@gmail.com>
|
2025-12-18 09:02:38 +00:00 |
|
|
Li, Jiang
|
e3ab93c896
|
[CPU] Refactor CPU fused MOE (#30531)
Signed-off-by: jiang1.li <jiang1.li@intel.com>
|
2025-12-18 14:36:49 +08:00 |
|
|
Vadim Gimpelson
|
717ac33d9c
|
[PERF] Qwen3-next. Add fp8 cutlass MoE tuned configs. chmod -x *MI308X.json (#29553)
Signed-off-by: Vadim Gimpelson <vadim.gimpelson@gmail.com>
|
2025-12-18 13:16:04 +08:00 |
|
|
Bowen Bao
|
0c738b58bc
|
[Quantization] Support Quark int4-fp8 w4a8 for MoE (#30071)
Signed-off-by: Bowen Bao <bowenbao@amd.com>
|
2025-12-18 04:20:42 +00:00 |
|
|
Varun Sundar Rabindranath
|
e3fc374a9a
|
[BugFix] Workspace allocation during profile run : DeepEPHighThroughput + DeepGEMM (#30899)
|
2025-12-17 15:00:59 -08:00 |
|