Commits · b7d1f17b8da9254a74687e63d576a63c4fb0cd10 · change / sglang · GitLab

08 Sep, 2025 17 commits
- Revert "enable auto-round quantization model (#6226)" (#10148) · b7d1f17b
  Yineng Zhang authored Sep 07, 2025
  
  b7d1f17b
- enable auto-round quantization model (#6226) · c8295d23
  Weiwei authored Sep 08, 2025
```
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
```
  c8295d23
- [Bugfix] Qwen3MoE aclrtMemcpy failed with NPUGraph (#10013) · b67c277f
  Even Zhou authored Sep 08, 2025
  
  b67c277f
- Fix: (glm4v) Add missing field (#10147) · 8116804e
  Xinyuan Tong authored Sep 08, 2025
  
  8116804e
- Add speculator attention backend switch (#9981) · 8c5930f0
  cicirori authored Sep 08, 2025
  
  8c5930f0
- [Bugfix] Retract not releasing enough memory when page size > 1 (#9989) · 3b99f23c
  Zhiqiang Xie authored Sep 07, 2025
  
  3b99f23c
- [1/N][Bug] Fix w4afp8 MoE NaN issue (sgl-kernel, fixed) (#10108) · ee0b3c5b
  Yuhao Yao authored Sep 08, 2025
  
  ee0b3c5b
- move compile threads to an option to avoid OOM on low memory host (#10123) · 6049ca20
  Rain Jiang authored Sep 07, 2025
  
  6049ca20
- Add graph runner support with torch compile on CPU (#7843) · 7577f0e4
  Cao E authored Sep 08, 2025
  
  7577f0e4
- Standalone speculative decoding (#10090) · 8cda5a62
  Qiaolin Yu authored Sep 07, 2025
  
  8cda5a62
- Fix run time error in dsv3-fp8 model on mi35x (#10104) · 400d3b97
  kk authored Sep 08, 2025
```
Co-authored-by: wunhuang <wunhuang@amd.com>
Co-authored-by: HaiShaw <hixiao@gmail.com>
Co-authored-by: Lianmin Zheng <lianminzheng@gmail.com>
```
  400d3b97
- Qwen2.5-VL eagle3 infer (#8801) · 37d83c6e
  Lzhang-hub authored Sep 08, 2025
  
  37d83c6e
- fix the fp8 topk_config.correction_bias is none bug (#10040) · 7802586c
  Rain Jiang authored Sep 07, 2025
  
  7802586c
- Fix slow fused add RMSNorm (#10141) · bc5fc332
  fzyzcjy authored Sep 08, 2025
  
  bc5fc332
- vlm: enable GLM4.1V server testing & fix video processing (#10095) · f3440adc
  Xinyuan Tong authored Sep 08, 2025
```
Signed-off-by: Xinyuan Tong <xinyuantong.cs@gmail.com>
Co-authored-by: Binyao Jiang <byjiang1996@gmail.com>
```
  f3440adc
- [MoE] fix: incorrect weight initialization for cutlass_fused_experts_fp8 (#10144) · 5a7e10fe
  Cheng Wan authored Sep 07, 2025
  
  5a7e10fe
- [BUG FIX] add fail check when get fail in case wait complete block (#9971) · 33467c05
  Shisong Ma authored Sep 08, 2025
```
Co-authored-by: mashisong <mashisong@bytedance.com>
Co-authored-by: Zhiqiang Xie <xiezhq@stanford.edu>
```
  33467c05
07 Sep, 2025 12 commits
- [DOC]: some minor updates (#10134) · b0fcbb74
  eigen authored Sep 07, 2025
  
  b0fcbb74
- Fix flashinfer version in sgl-kernel (#10135) · 76a2c86b
  Lianmin Zheng authored Sep 07, 2025
  
  76a2c86b
- [1/2] Refactor multi-tokenizer manager (#10074) · e719bb0e
  Liangsheng Yin authored Sep 07, 2025
  
  e719bb0e
- [Minor] fix lint in main (#10128) · 06724683
  DarkSharpness authored Sep 07, 2025
  
  06724683
- [Auto Sync] Update parallel_state.py (20250907) (#10126) · 617aa2b2
  Lianmin Zheng authored Sep 07, 2025
```
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: jzhou-xai <jzhou@x.ai>
```
  617aa2b2
- add dataset_path for bench_one_batch_server.py (#10113) · 111b1379
  miter authored Sep 07, 2025
```
Signed-off-by: linhuang <linhuang@ruijie.com.cn>
Co-authored-by: linhuang <linhuang@ruijie.com.cn>
```
  111b1379
- [HiCache] fix: check clear() method for storage backend (#10096) · 41628dc1
  Teng Ma authored Sep 07, 2025
```
Co-authored-by: hzh0425 <hzh0425@apache.org>
```
  41628dc1
- Fix cuda graph mode in flashinfer attn backend (#10056) · a12061df
  Ben Barsdell authored Sep 07, 2025
  
  a12061df
- Optimize nvfp4 block scaled gemm kernel when M is small. (#10101) · 85ed8e0a
  Qi Yuhang authored Sep 07, 2025
  
  85ed8e0a
- CUTLASS fp8 blockwise gemm support of sm120 (#9969) · dd1e2689
  Jianying authored Sep 07, 2025
  
  dd1e2689
- [Feature] LMCache Connector Integration (#9741) · 9a7ced4e
  Yuwei An authored Sep 06, 2025
```
Signed-off-by: Oasis-Git <ayw.sirius19@gmail.com>
Signed-off-by: YuhanLiu11 <yliu738@wisc.edu>
Co-authored-by: Zhiqiang Xie <xiezhq@stanford.edu>
```
  9a7ced4e
- Optimize moe_sum_reduce_kernel (#9477) · cb3918a0
  Yuan Luo authored Sep 07, 2025
```
Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>
Co-authored-by: Xiaoyu Zhang <35585791+BBuf@users.noreply.github.com>
```
  cb3918a0
06 Sep, 2025 11 commits
- [Auto Sync] Update server_args.py (20250906) (#10117) · f3b67602
  Lianmin Zheng authored Sep 06, 2025
```
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: Hanming Lu <69857889+hanming-lu@users.noreply.github.com>
```
  f3b67602
- [router] Improve the router e2e tests (#10102) · 9eb50ecc
  Keyang Ru authored Sep 06, 2025
  
  9eb50ecc
- increase the rust e2e timeout (#10116) · b3e7a2ce
  Keyang Ru authored Sep 06, 2025
  
  b3e7a2ce
- [CI] Refactor disaggregation tests (#10068) · 00974e4f
  Shangming Cai authored Sep 06, 2025
```
Signed-off-by: Shangming Cai <csmthu@gmail.com>
```
  00974e4f
- [chore] Remove unused ep_moe cuda kernels (#9956) · 5f1eb204
  hlu1 authored Sep 06, 2025
  
  5f1eb204
- Remove non-accelerated targets(100 and up) from cmake (#10041) · 039cef76
  hlu1 authored Sep 06, 2025
```
Signed-off-by: Hao Lu <14827759+hlu1@users.noreply.github.com>
```
  039cef76
- Disable kernel cutlass_mla_decode on SM103 (#10058) · 4c22ebe2
  hlu1 authored Sep 06, 2025
```
Signed-off-by: Hao Lu <14827759+hlu1@users.noreply.github.com>
```
  4c22ebe2
- Fix circular import (#10107) · a5a03209
  Cheng Wan authored Sep 06, 2025
  
  a5a03209
- [Fix] Compatibility between DP attention and pipeline parallelism (#10100) · 21af5c04
  Cheng Wan authored Sep 06, 2025
```
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
```
  21af5c04
- perf: Avoid unnecessary data type conversions for DeepSeek-V3 on Blackwell (#9834) · 012584ec
  Jinyang Yuan authored Sep 06, 2025
```
Signed-off-by: Jinyang Yuan <154768711+jinyangyuan-nvidia@users.noreply.github.com>
```
  012584ec
- [NVIDIA] disable chunked prefix cache when dp and blackwell is used (#9861) · 90dfe3de
  Kaixi Hou authored Sep 05, 2025
  
  90dfe3de