Commits · 2b7417bf6a9e2aa59a9736716f823fd03f26f67d · change / sglang · GitLab

20 Sep, 2025 8 commits
- fix(disagg): fix sending KV cache in case of MLA for NIXL backend (#10673) · 2b7417bf
  dmitrygx authored Sep 20, 2025
  
  2b7417bf
- Replace os.environ in layernorm.py (#10684) · f1116495
  Baizhou Zhang authored Sep 20, 2025
  
  f1116495
- [Performance] Qwen3-Next: optimize causal_conv1d_fn triton kernel - up to 9% faster (#10680) · bd7eb020
  Binyao Jiang authored Sep 20, 2025
  
  bd7eb020
- chore: upgrade mooncake 0.3.6.post1 to fix gb200 dockerfile (#10681) · 74cd6e39
  Shangming Cai authored Sep 20, 2025
```
Signed-off-by: Shangming Cai <csmthu@gmail.com>
```
  74cd6e39
- [Auto Sync] Update deepseek_v2.py (20250920) (#10683) · b17e67df
  Yineng Zhang authored Sep 19, 2025
```
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
```
  b17e67df
- [1/2] Support deterministic inference with flashinfer attention backend (#10645) · 8ecef73f
  Baizhou Zhang authored Sep 19, 2025
```
Co-authored-by: hebiao064 <hebiaobuaa@gmail.com>
Co-authored-by: Qiaolin-Yu <liin1211@outlook.com>
```
  8ecef73f
- [router] refactor router and worker management 2.5/n (#10677) · 1d1ce624
  Simo Lin authored Sep 19, 2025
  
  1d1ce624
- [Auto Sync] Update model_runner.py (20250920) (#10679) · 60e2a7ce
  Yineng Zhang authored Sep 19, 2025
```
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
```
  60e2a7ce
19 Sep, 2025 22 commits
- limit sgl-kernel causal conv1d to cuda only (#10648) · d88ef4a3
  Jinyan Chen authored Sep 20, 2025
```
Co-authored-by: Jinyan Chen <jinyanc@nvidia.com>
```
  d88ef4a3
- chore: cleanup docker image (#10671) · 6f993e8b
  Yineng Zhang authored Sep 19, 2025
  
  6f993e8b
- router-spec: Reorder `ChatCompletionRequest` and fix validation logic (#10675) · 03ce92e5
  Chang Su authored Sep 19, 2025
  
  03ce92e5
- [router] refactor router and worker management 2/n (#10666) · 00eb5eb7
  Simo Lin authored Sep 19, 2025
  
  00eb5eb7
- [Auto Sync] Update .clang-format (20250919) (#10670) · dab4663b
  Yineng Zhang authored Sep 19, 2025
```
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
```
  dab4663b
- fix: resolve sync issue (#10668) · 610a6d6e
  Yineng Zhang authored Sep 19, 2025
  
  610a6d6e
- [router] refactor router and worker management 1/n (#10664) · 36efd5be
  Simo Lin authored Sep 19, 2025
  
  36efd5be
- [router] preserve order of json params using preserve_order feature (#10661) · 68cdc189
  Fabian Gebhart authored Sep 19, 2025
  
  68cdc189
- [HiCacheStorage]support page_first_direct layout for generic set&get (#10522) · 7f399e4b
  huangtingwei authored Sep 19, 2025
  
  7f399e4b
- [router] refactor worker to builder pattern 5/n (#10653) · 873d858b
  Simo Lin authored Sep 19, 2025
  
  873d858b
- Fix fast decode plan for flashinfer v0.4.0rc1 and upgrade sgl-kernel 0.3.11 (#10634) · 3fa3c22a
  Baizhou Zhang authored Sep 19, 2025
```
Co-authored-by: zhyncs <me@zhyncs.com>
```
  3fa3c22a
- [router] refactor worker to builder pattern 4/n (#10650) · 4f2055ad
  Simo Lin authored Sep 19, 2025
  
  4f2055ad
- [sgl-kernel] Support moe_sum_reduce cuda kernel (#10321) · 616a3e20
  Yuan Luo authored Sep 19, 2025
```
Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>
Co-authored-by: Xiaoyu Zhang <35585791+BBuf@users.noreply.github.com>
```
  616a3e20
- [router] refactor worker to builder pattern 3/n (#10647) · ac2a723b
  Simo Lin authored Sep 19, 2025
  
  ac2a723b
- [Feature]feat(get_ip): unify get_ip_xxx (#10081) · 56b991b1
  Jimmy authored Sep 19, 2025
  
  56b991b1
- [router] refactor worker to builder pattern 2/n (#10633) · 780d6a22
  Simo Lin authored Sep 19, 2025
  
  780d6a22
- Hicache L3 backend mooncake optimization configuration reading method (#10319) · 8b713c72
  FlyPanda authored Sep 19, 2025
```
Co-authored-by: Teng Ma <sima.mt@alibaba-inc.com>
Co-authored-by: huangtingwei <141888744+huangtingwei9988@users.noreply.github.com>
Co-authored-by: shicang <shicang@shicang>
Co-authored-by: Shangming Cai <csmthu@gmail.com>
```
  8b713c72
- chore: bump sgl-kernel 0.3.11 (#10630) · 5bfafdfc
  Yineng Zhang authored Sep 18, 2025
  
  5bfafdfc
- feat: add fused moe config for Qwen3-Next-80B-A3B-Instruct on B200 (#10631) · 8c52de6f
  zixuanzhang226 authored Sep 18, 2025
  
  8c52de6f
- model support: Sarashina2VisionForCausalLM (#10632) · c1815a99
  Chang Su authored Sep 18, 2025
  
  c1815a99
- [Performance] Qwen3-Next: speed up update_mamba_state_after_mtp_verify by 10x;... · 4e6c4923
  Binyao Jiang authored Sep 18, 2025
```
[Performance] Qwen3-Next: speed up update_mamba_state_after_mtp_verify by 10x; e2e up to 3.54% faster (#10586)
```
  4e6c4923
- [Performance] Qwen3-Next: replace arange to cached query_start_loc_li… (#10553) · b91cb67e
  Binyao Jiang authored Sep 18, 2025
  
  b91cb67e
18 Sep, 2025 10 commits
- [Feature] Speculative decoding support lookahead (#9873) · e7bc6003
  Zhihao Zhang authored Sep 19, 2025
```
Co-authored-by: a4zhangfei <a4zhangfei@qq.com>
Co-authored-by: Qiaolin-Yu <liin1211@outlook.com>
```
  e7bc6003
- refactor: use registry for _get_attention_backend_from_str (#10629) · 2a2ff9a8
  Yineng Zhang authored Sep 18, 2025
  
  2a2ff9a8
- [router] refactor worker to builder pattern 1/n (#10628) · 5291f32d
  Simo Lin authored Sep 18, 2025
  
  5291f32d
- Garbage collector regression in the online server (#10621) · 67073dde
  brayden-hai authored Sep 18, 2025
  
  67073dde
- feat: Add FlexAttention Backend for Efficient Sparse Attention (#9947) · 9a5c42f9
  yuk.igalaxy authored Sep 19, 2025
```
Co-authored-by: Baizhou Zhang <sobereddiezhang@gmail.com>
```
  9a5c42f9
- Fix bias handling in TritonMoeQuantInfo within quantization/mxfp4.py (#10579) · 388c05d5
  yhyang201 authored Sep 19, 2025
  
  388c05d5
- [Performance] qwen3-next improve causal conv1d in prefill phase (#10595) · fc809665
  Jinyan Chen authored Sep 19, 2025
```
Co-authored-by: Jinyan Chen <jinyanc@nvidia.com>
```
  fc809665
- Fix sgl_kernel import failure on devices other than CUDA (#10610) · 6fd4816d
  Zaili Wang authored Sep 19, 2025
  
  6fd4816d
- support qwen3-next-fp8 deepep (#10622) · 1344ebc8
  Yi Zhang authored Sep 19, 2025
  
  1344ebc8
- update deepep version for qwen3-next deepep moe (#10624) · e07b21ce
  Yi Zhang authored Sep 19, 2025
  
  e07b21ce