Commits · 7a1f7fc5049d4fd5b3adc88fe48466fccbe676d8 · zhaoyu6 / sglang · GitLab

31 Jul, 2025 2 commits
- [Feature] Hybrid EP and TP (#8590) · 7a1f7fc5
  Cheng Wan authored Jul 31, 2025
  
  7a1f7fc5
- [4/N] MoE Refactor: Unified Triton Kernel for FusedMoE and EPMoE (#8515) · 32fa1e9c
  Cheng Wan authored Jul 31, 2025
  
  32fa1e9c
29 Jul, 2025 1 commit
- Support EPLB in FusedMoE (#8448) · 9effeb5b
  Cheng Wan authored Jul 29, 2025
  
  9effeb5b
28 Jul, 2025 3 commits
- Fix DEEPEP BF16 compatibility for Deepseek Style model like GLM 4.5 (#8469) · 74e7e457
  Stefan He authored Jul 28, 2025
```
Co-authored-by: Minglei Zhu <mingleizhu1122@gmail.com>
```
  74e7e457
- [3/N] MoE Refactor: Simplify DeepEP Output (#8421) · 9c138a04
  Cheng Wan authored Jul 28, 2025
  
  9c138a04
- [NVIDIA] Change to use `num_local_experts` (#8453) · 134fa43e
  Kaixi Hou authored Jul 28, 2025
  
  134fa43e
27 Jul, 2025 3 commits
- [hotfix] fix merge conflicts in FlashInferEPMoE (#8405) · 4d921f2b
  Cheng Wan authored Jul 27, 2025
  
  4d921f2b
- [2/N] MoE Refactor: Unify weight loader and quant methods (#8397) · bf0f448f
  Cheng Wan authored Jul 27, 2025
  
  bf0f448f
- [NVIDIA] Add Flashinfer MoE blockscale fp8 backend (#8036) · 85486b6f
  Kaixi Hou authored Jul 27, 2025
  
  85486b6f
22 Jul, 2025 1 commit
- [AMD] Remove vllm's scaled_fp8_quant and moe_sum when SGLANG_USE_AITER=1 (#7484) · e50109f2
  Hubert Lu authored Jul 21, 2025
  
  e50109f2
19 Jul, 2025 1 commit
- [1/N] MoE Refactor: refactor `select_experts` (#7966) · 15ad6c90
  Cheng Wan authored Jul 19, 2025
  
  15ad6c90
17 Jul, 2025 2 commits
- Refactor: move all quantization-related code to `srt/layer/quantization` (#7989) · 49b87774
  Cheng Wan authored Jul 17, 2025
  
  49b87774
- [Fix] ensure DeepGEMM is only enabled for FP8_W8A8 models (#8110) · 5c08a36c
  hzh0425 authored Jul 17, 2025
  
  5c08a36c
10 Jul, 2025 1 commit

[feature]Ascend quantization support (#7791) · 766392c6

ronnie_zheng authored Jul 10, 2025


Co-authored-by: ichernob <ichernobnn@gmail.com>
Co-authored-by: liupeng <liupeng374@huawei.com>

766392c6

07 Jul, 2025 1 commit
- feat: support DeepSeek-R1-W4AFP8 model with ep-moe mode (#7762) · cb9d91ea
  SijiaYang authored Jul 08, 2025
```
Signed-off-by: yangsijia.614 <yangsijia.614@bytedance.com>
```
  cb9d91ea
05 Jul, 2025 1 commit
- DP Attention with Auto DeepEP Dispatch (#7222) · 8fc910db
  Cheng Wan authored Jul 05, 2025
  
  8fc910db
03 Jul, 2025 1 commit

Ascend attention backend(PA&MLA) (#7722) · 1e0e5497

ronnie_zheng authored Jul 03, 2025


Co-authored-by: Maksim <makcum888e@mail.ru>
Co-authored-by: VDV1985 <vladdv85@mail.ru>

1e0e5497

29 Jun, 2025 1 commit
- Move files related to EPLB (#7580) · 0c9c6c75
  fzyzcjy authored Jun 30, 2025
  
  0c9c6c75
28 Jun, 2025 1 commit
- Let EP prefill support new DeepGEMM (#7310) · 00c7b1ad
  fzyzcjy authored Jun 28, 2025
  
  00c7b1ad
24 Jun, 2025 1 commit
- [AMD] add aiter fused moe in DeepEP path (#7268) · 755f3147
  Alex Sun authored Jun 24, 2025
  
  755f3147
23 Jun, 2025 1 commit

feat: integrate deepgemm into EPMoE (#6821) · 506c4928

xutizhou authored Jun 23, 2025


Co-authored-by: tianqilin.99 <tianqilin.99@bytedance.com>
Co-authored-by: TianQiLin666666 <1834987979@qq.com>
Co-authored-by: Cheng Wan <54331508+ch-wan@users.noreply.github.com>

506c4928

22 Jun, 2025 1 commit

FlashInfer NVFP4 MoE with EP & 2-stream shared expert (#7327) · 5962e70d

Trevor Morris authored Jun 22, 2025


Co-authored-by: JieXin Liang <Alcanderian@users.noreply.github.com>
Co-authored-by: alcanderian <alcanderian@gmail.com>

5962e70d

16 Jun, 2025 2 commits
- fix amd EP MoE FP8 issue (#7125) · 8e2363dc
  Alex Sun authored Jun 17, 2025
  
  8e2363dc
- [fix] fix DeepGEMM blackwell input quant & ut & fix style and log (#7247) · 5ca07eed
  JieXin Liang authored Jun 17, 2025
  
  5ca07eed
14 Jun, 2025 2 commits
- Support new DeepGEMM (#7172) · 93cec433
  fzyzcjy authored Jun 14, 2025
  
  93cec433
- Refactor DeepGEMM integration (#7150) · b4c41f72
  fzyzcjy authored Jun 14, 2025
  
  b4c41f72
05 Jun, 2025 1 commit
- Fix one missing arg in DeepEP (#6878) · 499f5e62
  Cheng Wan authored Jun 04, 2025
  
  499f5e62
04 Jun, 2025 1 commit
- Set `num_fused_shared_experts` as `num_shared_experts` when shared_experts... · 81964328
  Cheng Wan authored Jun 04, 2025
```
Set `num_fused_shared_experts` as `num_shared_experts` when shared_experts fusion is not disabled (#6736)
```
  81964328
31 May, 2025 1 commit
- Support token-level quantization for EP MoE (#6782) · ced3c07a
  Cheng Wan authored May 30, 2025
  
  ced3c07a
22 May, 2025 1 commit
- [RL] Remove the w13 weight_scale and input_scale for UnquantizedEPMoE… (#6308) · e9feb488
  Zilin Zhu authored May 22, 2025
  
  e9feb488
20 May, 2025 3 commits
- Fix master CI for DeepSeek (#6447) · 13feffd0
  fzyzcjy authored May 20, 2025
  
  13feffd0
- Support dispatching logical to physical experts (#6385) · e98afbe0
  fzyzcjy authored May 20, 2025
  
  e98afbe0
- Support loading weights when physical experts are different from logical experts (#6386) · c471d39e
  fzyzcjy authored May 20, 2025
  
  c471d39e
17 May, 2025 1 commit
- Minor code cleanup refactor for DeepSeek models (#6324) · 2df9d40a
  fzyzcjy authored May 17, 2025
  
  2df9d40a
15 May, 2025 1 commit
- Reduce MoE memory usage (#6147) · f194e14f
  fzyzcjy authored May 16, 2025
  
  f194e14f
11 May, 2025 1 commit
- Add typo checker in pre-commit (#6179) · 2ce87935
  applesaucethebun authored May 11, 2025
```
Co-authored-by: Brayden Zhong <b8zhong@uwaterloo.ca>
```
  2ce87935
08 May, 2025 1 commit

DeepEP normal support deepgemm-contiguous (#5626) · acc816d8

lukec authored May 08, 2025


Co-authored-by: Yingyi Huang <yingyihuang2000@outlook.com>
Co-authored-by: Cheng Wan <54331508+ch-wan@users.noreply.github.com>
Co-authored-by: Xuting Zhou <xutingz@nvidia.com>
Co-authored-by: ZhengHSI <zhenghsi@qq.com>

acc816d8

20 Apr, 2025 1 commit
- Fix DeepEP cannot run on latest master (#5567) · 463d4b74
  fzyzcjy authored Apr 21, 2025
  
  463d4b74
19 Apr, 2025 2 commits
- simplify the control logic for using shared experts fusion (#5504) · d58e3544
  Xiaoyu Zhang authored Apr 20, 2025
  
  d58e3544
- Fix DeepGEMM masked cannot be run on groups not being multiple or 4 (#5340) · 1e0806f3
  fzyzcjy authored Apr 19, 2025
  
  1e0806f3