Commits · 22fa63cfe849c38bbdb590163043230a1a1a6148 · OpenDAS / vllm_cscc

22 Apr, 2026 14 commits

[Bugfix][Torch 2.12] Fix batch_invariant test with allow_override for torch 2.12 upgrade (#40562) · 22fa63cf
Lucas Kabela authored Apr 22, 2026
```
Signed-off-by: Lucas Kabela <lucaskabela@meta.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
```
22fa63cf

[Refactor] Clean up log once `scope="local"` (#40540) · 8f87eb46

Wentao Ye authored Apr 22, 2026

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>

8f87eb46

[NVFP4] NVFP4 MOE emulation fallback for H100/MI300/MI350, standardize... · d622e27d

fxmarty-amd authored Apr 22, 2026


[NVFP4] NVFP4 MOE emulation fallback for H100/MI300/MI350, standardize `TritonExperts` usage for OCP MX emulation (#35737)
Signed-off-by: Felix Marty <Felix.Marty@amd.com>
Signed-off-by: fxmarty-amd <felmarty@amd.com>
Co-authored-by: Kyle Sayers <kylesayrs@gmail.com>

d622e27d

[MoE] Convert CT W8A8 To Oracle Structure (#39187) · 5f76b3fb

Robert Shaw authored Apr 22, 2026


Signed-off-by: Robert Shaw <robertgshaw2@gmail.com>
Co-authored-by: Claude <noreply@anthropic.com>

5f76b3fb

[MoE Refactor] Combine MoERunnerBase + DefaultMoERunner (#40560) · 809d83c2
bnellnm authored Apr 22, 2026
```
Signed-off-by: Bill Nell <bnell@redhat.com>
```
809d83c2
[xpu][rocm] Update `current_platform.supports_fp8()` for TritonExperts (#40132) · a4905133
Hank_ authored Apr 22, 2026
```
Signed-off-by: Hank <hcc.mayday@gmail.com>
```
a4905133

[Bugfix] LoRA for DeepSeek V3.2 (#35077) · a250f1bd

ℍ𝕠𝕝𝕝𝕠𝕨 𝕄𝕒𝕟 authored Apr 22, 2026


Signed-off-by: Hollow Man <hollowman@opensuse.org>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
Co-authored-by: Jee Jee Li <pandaleefree@gmail.com>

a250f1bd

support hotwords for FunASR model (#39674) · 9047288b

AllenDou authored Apr 22, 2026


Signed-off-by: zixiao <shunli.dsl@alibaba-inc.com>
Co-authored-by: zixiao <shunli.dsl@alibaba-inc.com>

9047288b

[Model] Add block-local attention and YaRN for local layers to Gemma3 (#39823) · 12367487
philip-essential authored Apr 21, 2026
```
Signed-off-by: Philip Monk <169196560+philip-essential@users.noreply.github.com>
```
12367487

[Audio] Bundle `get_generation_prompt()` params into `SpeechToTextParams` (#36268) · 4eafc729

Ekagra Ranjan authored Apr 22, 2026

Signed-off-by: Ekagra Ranjan <3116519+ekagra-ranjan@users.noreply.github.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com>

4eafc729

[compile] mla + group fp8 fusion (#38877) · 4506319a

Carl Y authored Apr 21, 2026


Signed-off-by: Carl You <4531192+carlyou@users.noreply.github.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>

4506319a

[Bugfix] Fix quantized model initialization failure with prefetch offloading (#40432) · 9b60e2ff

Rishapveer Singh authored Apr 22, 2026


Signed-off-by: Rishapveer Singh <singhrishapveer@gmail.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>

9b60e2ff

[MyPy] Enable mypy for `vllm/model_executor/layers/` (#40159) · 3951d3ea
Martin Hickey authored Apr 22, 2026
```
Signed-off-by: Martin Hickey <martin.hickey@ie.ibm.com>
```
3951d3ea

[Bugfix] Fix W4A8_FP8 MoE tp>1 correctness and view() TypeError (#40310) · f946659f

EdalatiAli authored Apr 21, 2026

Signed-off-by: EdalatiAli <aliedalati@cohere.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>

f946659f

21 Apr, 2026 12 commits
- [MoE Refactor] Add more MoE layer tests (#39349) · 9db4650e
  bnellnm authored Apr 21, 2026
```
Signed-off-by: Bill Nell <bnell@redhat.com>
Signed-off-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
```
  9db4650e
- [MoE Refactor] Remove SharedFusedMoE class (#35782) · 5e584ce9
  bnellnm authored Apr 21, 2026
```
Signed-off-by: Bill Nell <bnell@redhat.com>
```
  5e584ce9
- [Refactor] Remove unused param (#39750) · 1842447c
  Wentao Ye authored Apr 21, 2026
```
Signed-off-by: yewentao256 <zhyanwentao@126.com>
```
  1842447c
- [Perf] Optimize batch invariant with fused rms norm, 2.1% E2E latency improvement (#40413) · 16688b26
  Wentao Ye authored Apr 21, 2026
```
Signed-off-by: yewentao256 <zhyanwentao@126.com>
```
  16688b26
- Add new tp plan styles to the Transformers modelling backend (#40467) · 6ee081d1
  Harry Mellor authored Apr 21, 2026
```
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
```
  6ee081d1
- Default to 'align' mamba cache mode for Mamba-based models when speculative... · f819265a
  roikoren755 authored Apr 21, 2026
```
Default to 'align' mamba cache mode for Mamba-based models when speculative decoding is enabled (#40454)
Signed-off-by: Roi Koren <roik@nvidia.com>
```
  f819265a
- [MM][CG] Optimize default `max_frames_per_batch` auto-infer for ViT CUDA graph... · 936e0b79
  Shanshan Shen authored Apr 21, 2026
```
[MM][CG] Optimize default `max_frames_per_batch` auto-infer for ViT CUDA graph video inference (#40445)
Signed-off-by: shen-shanshan <467638484@qq.com>
```
  936e0b79
- [Bugfix] LoRA: extend expert base_layer loading to Qwen3.5 and Step3.x (#37114) · 908a7134
  ℍ𝕠𝕝𝕝𝕠𝕨 𝕄𝕒𝕟 authored Apr 21, 2026
```
Signed-off-by: Hollow Man <hollowman@opensuse.org>
```
  908a7134
- Add Granite 4.1 Vision as built-in multimodal model (#40282) · d249a9e9
  artem-spector authored Apr 21, 2026
```
Signed-off-by: Artem Spector <artems@il.ibm.com>
Signed-off-by: artemspector <artems@il.ibm.com>
Co-authored-by: artemspector <artems@il.ibm.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
```
  d249a9e9
- [MoE] Triton MoE Perf regression - restore low latency path (#39016) · 257015d5
  milesial authored Apr 20, 2026
  
  257015d5
- [Bugfix] Gemma4: fix multimodal embedder norm order to match HF reference (#40411) · 20d37434
  Luciano Martins authored Apr 20, 2026
```
Signed-off-by: Luciano Martins <lucianommartins@users.noreply.github.com>
Co-authored-by: Luciano Martins <lucianommartins@users.noreply.github.com>
```
  20d37434
- [Misc] Reduce attention logging levels (#40086) · 18563f20
  Chauncey authored Apr 21, 2026
```
Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com>
```
  18563f20
20 Apr, 2026 10 commits
- [Bugfix] Replace code that disabled shared expert overlap (#39222) · 6867bcd0
  bnellnm authored Apr 20, 2026
```
Signed-off-by: Bill Nell <bnell@redhat.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
```
  6867bcd0
- Update flashinfer to 0.6.8 (#39959) · 191e3fda
  bai authored Apr 20, 2026
```
Signed-off-by: bai <v@gor.io>
```
  191e3fda
- [MoE Refactor] Move the shared/fused expert output sum into MoERunnerBase (#35949) · 726efe17
  bnellnm authored Apr 20, 2026
```
Signed-off-by: Bill Nell <bnell@redhat.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
```
  726efe17
- [XPU] fix MoE triton backend in online fp8 quantization (#40109) · 59556265
  Yan Ma authored Apr 20, 2026
```
Signed-off-by: Yan Ma <yan.ma@intel.com>
```
  59556265
- Properly enable wvSplitK fp8 path for RDNA (#37712) · 3a30eaa1
  Hashem Hashemi authored Apr 20, 2026
```
Signed-off-by: Hashem Hashemi <hashem.hashemi@amd.com>
```
  3a30eaa1
- mxfp8 online quant move to new frontend (#40152) · 38fa87ca
  Vasiliy Kuznetsov authored Apr 20, 2026
```
Signed-off-by: Vasiliy Kuznetsov <vasiliy@meta.com>
```
  38fa87ca
- [Bugfix] Fix scaled_mm output narrowing for 3D input tensors (#38093) · 58631d7c
  nemanjaudovic authored Apr 20, 2026
```
Signed-off-by: nemanjaudovic <nudovic@amd.com>
```
  58631d7c
- [BUGFIX] Fix Pixtral consolidated format vision weight loading (#39916) · 6097afb9
  Julien Denize authored Apr 20, 2026
```
Signed-off-by: Julien Denize <julien.denize@mistral.ai>
Signed-off-by: juliendenize <julien.denize@mistral.ai>
```
  6097afb9
- [Qwen][Bugfix] Fixes sigmoid activation in torch impl of RMSNormGated. (#40245) · 89361181
  Tao He authored Apr 20, 2026
```
Signed-off-by: Tao He <linzhu.ht@alibaba-inc.com>
```
  89361181
- [Bugfix] Properly initialize `PerTensorScaleParameter` for fused-on-disk checkpoints (#39765) · fcb31c1a
  Shinichi Hemmi authored Apr 20, 2026
```
Signed-off-by: Hemmi Shinichi <shemmi@preferred.jp>
Signed-off-by: Shinichi Hemmi <50256998+Alnusjaponica@users.noreply.github.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
```
  fcb31c1a
19 Apr, 2026 2 commits
- Fix MoE backend selection for LoRA (unquantized MoE) (#40273) · d1135a50
  danisereb authored Apr 19, 2026
```
Signed-off-by: Daniel Serebrenik <daserebrenik@nvidia.com>
```
  d1135a50
- [FEAT] [Perf] [Gemma4] Fused Gemma4 Routing Function Triton (#39083) · 45232a45
  TJian authored Apr 19, 2026
```
Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>
```
  45232a45
18 Apr, 2026 2 commits
- [Attention] TurboQuant: remove redundant random signs, add prior art attribution (#40194) · ed0622e3
  Dan Alistarh authored Apr 18, 2026
```
Signed-off-by: Dan Alistarh <d.alistarh@gmail.com>
```
  ed0622e3
- Added general ND x ND matmul and unit test for it (#39909) · b5f6c5f8
  Yusuf Mohammad authored Apr 18, 2026
```
Signed-off-by: Yusuf <yusufmohammad@live.com>
```
  b5f6c5f8