Commits · 01923eec7092fd5b718cb9b45eb6df152abe9296 · OpenDAS / vllm_cscc

26 Feb, 2026 16 commits
- [ROCm][Quantization] GPT OSS Upstream MoE wmxfp4_afp8 with static scales (#30357) · 01923eec
  Aleksandr Malyshev authored Feb 26, 2026
```
Signed-off-by: Aleksandr Malyshev <maleksan@amd.com>
Co-authored-by: Aleksandr Malyshev <maleksan@amd.com>
```
  01923eec
- [WideEP] Remove pplx all2all backend (#33724) · eb19955c
  Tyler Michael Smith authored Feb 26, 2026
```
Signed-off-by: Tyler Michael Smith <tlrmchlsmth@gmail.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
```
  eb19955c
- add mixed precision support for modelopt (#35047) · d0105b84
  sychen52 authored Feb 26, 2026
```
Signed-off-by: Shiyang Chen <shiychen@nvidia.com>
```
  d0105b84
- Nemotron: use per-layer config in NemotronHMLPDecoderLayer for heterogeneous models (#35396) · 832a780f
  danielafrimi authored Feb 26, 2026
```
Signed-off-by: dafrimi <dafrimi@nvidia.com>
```
  832a780f
- [ROCm] Add dynamic mxfp4 quantization for DeepSeek V2 projection layers (#34157) · ec8ab9d2
  Douglas Lehr authored Feb 26, 2026
```
Signed-off-by: Doug Lehr <douglehr@amd.com>
Signed-off-by: Douglas Lehr <91553416+dllehr-amd@users.noreply.github.com>
Co-authored-by: Doug Lehr <douglehr@amd.com>
Co-authored-by: Rohan Potdar <66227218+Rohan138@users.noreply.github.com>
Co-authored-by: Gregory Shtrasberg <156009573+gshtras@users.noreply.github.com>
```
  ec8ab9d2
- [Refactor] Remove dead or duplicate func utils or variables (#35318) · 05972ea7
  Wentao Ye authored Feb 26, 2026
```
Signed-off-by: yewentao256 <zhyanwentao@126.com>
```
  05972ea7
- [Model] Add nvidia/llama-nemotron-embed-vl-1b-v2 multimodal embedding model (#35297) · 111d8690
  Jakub Zakrzewski authored Feb 26, 2026
```
Signed-off-by: Jakub Zakrzewski <jzakrzewski@nvidia.com>
```
  111d8690
- [Misc] Standardize handling of `mm_processor_kwargs.size` (#35284) · 845ee348
  Cyrus Leung authored Feb 26, 2026
```
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
```
  845ee348
- [Bugfix] fix device_name for routing replay (#34336) · c6ca5159
  Li-Yongwen authored Feb 26, 2026
```
Signed-off-by: liyongwen <1310439159@qq.com>
```
  c6ca5159
- [Bugfix] Fix Qwen2.5-Omni and Qwen3-Omni mixed-modality embed regression (#35368) · c0615a29
  Yueqian Lin authored Feb 26, 2026
```
Signed-off-by: linyueqian <linyueqian@outlook.com>
```
  c0615a29
- [Bugfix] [Qwen3.5]Fix Qwen3.5 FP8 quantization: tuple shard_id weight loading (#35289) · 32693db8
  HZY authored Feb 26, 2026
```
Signed-off-by: daowu.hzy <daowu.hzy@alibaba-inc.com>
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
Co-authored-by: Isotr0py <mozf@mail2.sysu.edu.cn>
```
  32693db8
- [Model] Ring 2.5 (#35102) · ab87f852
  Jiangyun Zhu authored Feb 26, 2026
```
Signed-off-by: zjy0516 <riverclouds.zhu@qq.com>
```
  ab87f852
- [CPU][Feat] Enable KleidiAI INT8_W4A8 for all input dtypes (#34890) · 4171ff6d
  Fadi Arafeh authored Feb 26, 2026
```
Signed-off-by: Fadi Arafeh <fadi.arafeh@arm.com>
Co-authored-by: Tyler Michael Smith <tlrmchlsmth@gmail.com>
```
  4171ff6d
- [UX] Add `--moe-backend` arg for explicit kernel selection (#33807) · de527e1c
  Michael Goin authored Feb 25, 2026
```
Signed-off-by: mgoin <mgoin64@gmail.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
```
  de527e1c
- [MoE Refactor] MXFP4 Cutlass Experts to MK (#34542) · 1976356e
  Yongye Zhu authored Feb 25, 2026
```
Signed-off-by: Yongye Zhu <zyy1102000@gmail.com>
```
  1976356e
- [offloader] v2: Hide weight onloading latency via prefetching (#29941) · 6831650c
  Ming Yang authored Feb 25, 2026
```
Signed-off-by: Ming Yang <minos.future@gmail.com>
Signed-off-by: Michael Goin <mgoin64@gmail.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
```
  6831650c
25 Feb, 2026 7 commits
- fix(mxfp4): Disable monolithic path for TRITON backend with EP (#34270) · c97234c0
  Elizabeth Thomas authored Feb 25, 2026
```
Signed-off-by: Elizabeth Thomas <email2eliza@gmail.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
```
  c97234c0
- Revert "[Misc] Enable weights loading tracking for quantized models" (#35309) · 15d76f74
  Lucas Wilkinson authored Feb 25, 2026
  
  15d76f74
- [Perf] Add opt-in SM100 Oink RMSNorm custom-op path (#31828) · 24650715
  Laura Wang authored Feb 24, 2026
```
Signed-off-by: Laura Wang <3700467+Laurawly@users.noreply.github.com>
Co-authored-by: Lu Fang <30275821+houseroad@users.noreply.github.com>
```
  24650715
- [Platform] Add current_platform.num_compute_units interface (#35042) · 8ad54a99
  Kunshang Ji authored Feb 25, 2026
```
Signed-off-by: Kunshang Ji <kunshang.ji@intel.com>
Signed-off-by: Kunshang Ji <jikunshang95@gmail.com>
```
  8ad54a99
- [Misc] Add shard_id validation for MergedColumnLinear (#35055) · a6c13752
  Isotr0py authored Feb 25, 2026
```
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
```
  a6c13752
- [Misc] Enable weights loading tracking for quantized models (#35074) · 4572a06a
  Isotr0py authored Feb 25, 2026
```
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
```
  4572a06a
- [ROCm]: Enable customop and rope+kvcache fusion for AITER RoPE (#35180) · f38f8c97
  Rohan Potdar authored Feb 24, 2026
```
Signed-off-by: Rohan138 <rohanpotdar138@gmail.com>
```
  f38f8c97
24 Feb, 2026 13 commits

Adding Nemotron fp8 Triton MoE Config (#34674) · 576fe503

yugong333 authored Feb 24, 2026


Signed-off-by: Yu Gong <yu3.gong@gmail.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>

576fe503

[Model][Spec Decode] Nemotron-H MTP and Mamba Speculative Decoding Support (#33726) · f5972a87

Benjamin Chislett authored Feb 24, 2026


Signed-off-by: Shahar Mor <smor@nvidia.com>
Signed-off-by: Benjamin Chislett <bchislett@nvidia.com>
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Co-authored-by: Shahar Mor <smor@nvidia.com>
Co-authored-by: Roi Koren <roik@nvidia.com>
Co-authored-by: Lucas Wilkinson <lwilkins@redhat.com>

f5972a87

Remove `padding_index` from models that don't use it for better Transformers... · c38b8d5a

Harry Mellor authored Feb 24, 2026


Remove `padding_index` from models that don't use it for better Transformers v5 compatibility (#35189)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>

c38b8d5a

Integrate flashinfer mm_mxfp8 in ModelOpt MXFP8 (#35053) · 9609b1f1
danisereb authored Feb 24, 2026
```
Signed-off-by: Daniel Serebrenik <daserebrenik@nvidia.com>
```
9609b1f1
Fix fallback to default tactic (flashinfer autotuner) with trtllm_fp4_block_scale_moe (#35088) · a0c70816
danisereb authored Feb 24, 2026
```
Signed-off-by: Daniel Serebrenik <daserebrenik@nvidia.com>
```
a0c70816

[Attn,KV-cache] Use per-head scales in the attention selector (#34281) · a87cc508

Eldar Kurtić authored Feb 24, 2026


Signed-off-by: Your Name <you@example.com>
Signed-off-by: Eldar Kurtic <research@neuralmagic.com>
Co-authored-by: Eldar Kurtic <research@neuralmagic.com>
Co-authored-by: Your Name <you@example.com>

a87cc508

[Feature] Add LoRA tower/connector support for Llama 4 Vision (mllama4) (#35147) · 012dee92
Dor Huri authored Feb 24, 2026
```
Signed-off-by: dorhuri123 <dor.huri1@live.biu.ac.il>
Co-authored-by: Jee Jee Li <pandaleefree@gmail.com>
```
012dee92

Make voxtral compile friendly (#33959) · f1c66454

Tugsbayasgalan Manlaibaatar authored Feb 24, 2026


Signed-off-by: Tugsbayasgalan Manlaibaatar <tmanlaibaatar@fb.com>
Co-authored-by: Nicolò Lucchesi <nlucches@redhat.com>

f1c66454

[Refactor] [1/N] Reorganize kernel abstraction directory (#34055) · 6af03f23

BadrBasowid authored Feb 24, 2026


Signed-off-by: BadrBasowid <badr.basowid@gmail.com>
Co-authored-by: vllmellm <vllm.ellm@embeddedllm.com>
Co-authored-by: TJian <tunjian.tan@embeddedllm.com>

6af03f23

[BUGFIX][Qwen3.5] Hardcode `mlp.gate` as not quantizable (#35156) · 33a0d43c
Vadim Gimpelson authored Feb 24, 2026
```
Signed-off-by: Vadim Gimpelson <vadim.gimpelson@gmail.com>
```
33a0d43c

[Quantization] Support FP8 MoE bias for models like GPT-OSS (#34906) · ec853405

Jia Guo authored Feb 23, 2026


Signed-off-by: jasperjiaguo <jasperg662@gmail.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

ec853405

[ROCm] AITER fused RoPE+KVCache (#33443) · 2ff4e511

Rohan Potdar authored Feb 23, 2026


Signed-off-by: Rohan138 <rohanpotdar138@gmail.com>
Signed-off-by: charlifu <charlifu@amd.com>
Signed-off-by: Rohan Potdar <66227218+Rohan138@users.noreply.github.com>
Co-authored-by: charlifu <charlifu@amd.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
Co-authored-by: Douglas Lehr <91553416+dllehr-amd@users.noreply.github.com>

2ff4e511

[Perf] Improve default triton fused moe configs (#34846) · 22a97e66

Michael Goin authored Feb 23, 2026


Signed-off-by: mgoin <mgoin64@gmail.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>

22a97e66

23 Feb, 2026 4 commits
- Use Xet high performance mode for Transformers v5 (#35098) · c4f38696
  Harry Mellor authored Feb 23, 2026
```
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
```
  c4f38696
- [Bugfix] Fix MRotaryEmbedding missing `truncate` attr with YaRN scaling (#35080) · a7f341c3
  haosdent authored Feb 24, 2026
```
Signed-off-by: haosdent <haosdent@gmail.com>
```
  a7f341c3
- [kv-cache, ct] Use compressed-tensors as a source of ground-truth for quant strategies (#34254) · b95bb692
  Eldar Kurtić authored Feb 23, 2026
```
Signed-off-by: Your Name <you@example.com>
Co-authored-by: Your Name <you@example.com>
```
  b95bb692
- [Refactor] Decouple TimingContext from InputProcessingContext (#35083) · 39264545
  Cyrus Leung authored Feb 23, 2026
```
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
```
  39264545