Commits · 298509008451f7861a848d477829d5816eef12cd · change / sglang · GitLab

05 Sep, 2025 16 commits
- Update flashinfer to 0.3.1 for B300 support (#10087) · 29850900
  hlu1 authored Sep 05, 2025
```
Signed-off-by: Hao Lu <14827759+hlu1@users.noreply.github.com>
```
  29850900
- [bugfix]: use correct cache location for cross attention in torch native backend (#8622) · e678cc71
  Mahmoud Ashraf authored Sep 05, 2025
  
  e678cc71
- enable aiter gemm_a8w8_bpreshuffle for ptpc gemm (#8555) · 4efe844a
  Morpheus Guo authored Sep 06, 2025
  
  4efe844a
- Tiny let DeepGEMM scale checks cover more cases (#7182) · 5e5c30d9
  fzyzcjy authored Sep 05, 2025
```
Co-authored-by: Yineng Zhang <me@zhyncs.com>
```
  5e5c30d9
- Fix and enhance dumper (#8725) · 9f00ec44
  fzyzcjy authored Sep 05, 2025
  
  9f00ec44
- Support simple evals in text comparator (#8867) · 8e85ee88
  fzyzcjy authored Sep 05, 2025
  
  8e85ee88
- Forbid DeepEP racing condition when too many tokens (#9567) · adf73175
  fzyzcjy authored Sep 05, 2025
  
  adf73175
- [Fix] Add speculative_draft_model_revision to server_args (#5255) · 13705dae
  DevashishLal-CB authored Sep 05, 2025
```
Signed-off-by: Devashish Lal <devashish@rivosinc.com>
```
  13705dae
- Tiny support setting numa nodes for different ranks (#10006) · df97b31f
  fzyzcjy authored Sep 05, 2025
  
  df97b31f
- [1/2] Optimizations and refactors about quant kernel (#9534) · 339f8eef
  fzyzcjy authored Sep 05, 2025
  
  339f8eef
- Fix typo in scheduler (#9934) · afd9f2f5
  limingshu authored Sep 05, 2025
  
  afd9f2f5
- [Vulnerability]feat(conn): set bootstrap server host (#9931) · f40038fb
  Jimmy authored Sep 05, 2025
  
  f40038fb
- Integrate trtllm ragged attention for prefill self-attention (#9801) · bebd0576
  Elfie Guo authored Sep 05, 2025
  
  bebd0576
- fix MultiTokenizerWrapper name (#10049) · f9836660
  Huang Long authored Sep 05, 2025
```
Signed-off-by: huanglong <huanglong@linux.alibaba.com>
```
  f9836660
- Simplify `Router` arguments passing and build it in docker image (#9964) · 6e95f5e5
  Liangsheng Yin authored Sep 05, 2025
  
  6e95f5e5
- chore: bump v0.5.2rc2 (#10050) · fa9c82d3
  Yineng Zhang authored Sep 04, 2025
  
  fa9c82d3
04 Sep, 2025 14 commits
- Fix accuracy drop of dsv3 run in dp enablement (#8677) · 918e3d4c
  kk authored Sep 05, 2025
```
Co-authored-by: wunhuang <wunhuang@amd.com>
```
  918e3d4c
- Optimized deepseek-v3/r1 model performance on mxfp4 run (#10008) · e9697374
  kk authored Sep 05, 2025
```
Co-authored-by: wunhuang <wunhuang@amd.com>
Co-authored-by: HAI <hixiao@gmail.com>
Co-authored-by: Hubert Lu <55214931+hubertlu-tw@users.noreply.github.com>
```
  e9697374
- [Hicache] Mooncake API Fix & Test, and Improved Readme (#9951) · 93088b69
  ykwd authored Sep 05, 2025
```
Co-authored-by: Teng Ma <sima.mt@alibaba-inc.com>
```
  93088b69
- Save memory for expert model parallel (#9957) · 453511ac
  Cheng Wan authored Sep 04, 2025
  
  453511ac
- fix 3fs zerocopy (#9938) · d0730487
  pansicheng authored Sep 05, 2025
```
Co-authored-by: Zhiqiang Xie <xiezhq@stanford.edu>
```
  d0730487
- metrics: support customer buckets for prompt/generation_tokens_histogram (#9634) · b32ab070
  Yingchun Lai authored Sep 04, 2025
  
  b32ab070
- [Doc] Fix SGLang tool parser doc (#9886) · 75ee0011
  Huapeng Zhou authored Sep 04, 2025
  
  75ee0011
- Optimize Qwen3-moe model by using flashinfer fused allreduce (#9973) · ec15c836
  Yuan Luo authored Sep 04, 2025
```
Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>
```
  ec15c836
- feat(hicache): Add generic hicache ci e2e test and benchmark test (#9846) · 106c2b31
  hzh0425 authored Sep 04, 2025
```
Co-authored-by: Zhiqiang Xie <xiezhq@stanford.edu>
```
  106c2b31
- Ensure chunked request extension length respects both rem_chunk_tokens and... · c6756949
  pansicheng authored Sep 04, 2025
```
Ensure chunked request extension length respects both rem_chunk_tokens and rem_total_tokens limits (#10003)
```
  c6756949
- [1/N] DP-refactor: move dp balance code into scheduler's mixin class (#10004) · 27e8ffed
  Liangsheng Yin authored Sep 04, 2025
  
  27e8ffed
- fix: health_generate endpoint in mini_lb (#9997) · 4dbb34fe
  wxsm authored Sep 04, 2025
  
  4dbb34fe
- [Bugfix] fix pd chat completion protocol for batching support (#10016) · 1e18a341
  Tony Lu authored Sep 04, 2025
```
Signed-off-by: Tony Lu <tonylu@linux.alibaba.com>
```
  1e18a341
- Fix Llama 4 with MXFP4 dynamic quant on MI35x (#9993) · 2c562fd2
  Hubert Lu authored Sep 04, 2025
  
  2c562fd2
03 Sep, 2025 10 commits
- [Auto Sync] Update parallel_state.py, few_shot_gsm8k.py (20250903) (#9986) · 397448eb
  Lianmin Zheng authored Sep 03, 2025
```
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: Leon Gao <leon.gao19@gmail.com>
```
  397448eb
- Minor update regarding issue #9704 (#9733) · 66d5d042
  Elfie Guo authored Sep 03, 2025
  
  66d5d042
- Triton 3.4.0 MoE config for Deepseek TP16 H100 (#9978) · 8cbf71dc
  Szymon Ożóg authored Sep 03, 2025
  
  8cbf71dc
- fix swa clear(): rename is_in_free_group to is_not_in_free_group (#9914) · 56eb5d0a
  Xinyuan Tong authored Sep 03, 2025
  
  56eb5d0a
- Remove mrope position sync (#9460) · 4ed9053e
  timmy-feng authored Sep 03, 2025
```
Co-authored-by: Nathan Wang <nathan.r.wang@gmail.com>
```
  4ed9053e
- Revert "Optimized deepseek-v3/r1 model performance on mxfp4 run (#9671)" (#9959) · 1b2ff4fb
  Yineng Zhang authored Sep 03, 2025
  
  1b2ff4fb
- Revert "[Fix] DeepSeek EP accuracy issue on B200 GPUs (#9946)" (#9955) · 2c7ca33a
  Yineng Zhang authored Sep 02, 2025
  
  2c7ca33a
- [feat] Add P/D attention select for draft model (#9755) · df397a72
  Ximingwang-09 authored Sep 03, 2025
```
Co-authored-by: 纬杭 <ximing.wxm@antgroup.com>
```
  df397a72
- add proctitle for tokenizers (#9952) · 5dfcd6c2
  Liangsheng Yin authored Sep 03, 2025
```
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
```
  5dfcd6c2
- Optimized deepseek-v3/r1 model performance on mxfp4 run (#9671) · 0dfd54d1
  kk authored Sep 03, 2025
```
Co-authored-by: wunhuang <wunhuang@amd.com>
Co-authored-by: wghuang <wghuang@amd.com>
```
  0dfd54d1