Commits · d658f0497efcfd5c256967d4f9d6505bf2c4e702 · change / sglang · GitLab

19 Oct, 2025 16 commits
- [overlap-spec] fix stop condition and trimming (#11819) · d658f049
  Liangsheng Yin authored Oct 19, 2025
  
  d658f049
- Revert "Fix: Dynamic RoPE Cache Expansion to Prevent Position-ID Out-of-Bounds... · 57e25de7
  Liangsheng Yin authored Oct 19, 2025
```
Revert "Fix: Dynamic RoPE Cache Expansion to Prevent Position-ID Out-of-Bounds in EAGLE + Long-Sequence Workloads" (#11827)
```
  57e25de7
- Change bf16 to fp8 for some gemms in attention for DeepSeek ckpt v2 (#11805) · 12eb02e9
  fzyzcjy authored Oct 19, 2025
  
  12eb02e9
- Avoid generation gets hanging when user specifies multiple event loops (#5162) · 002d0373
  fzyzcjy authored Oct 19, 2025
```
Co-authored-by: Yineng Zhang <me@zhyncs.com>
```
  002d0373
- Support not officially supported high sgl-kernel version with low srt version (#11786) · a27825ae
  fzyzcjy authored Oct 19, 2025
  
  a27825ae
- Make single-batch overlap compatible with NextN (#11804) · ce399e15
  fzyzcjy authored Oct 19, 2025
  
  ce399e15
- Tiny add hints when users send requests to wrong place (#11808) · ea6275df
  fzyzcjy authored Oct 19, 2025
  
  ea6275df
- support tokenized batch request (#11091) · eb7318f1
  narutolhy authored Oct 19, 2025
  
  eb7318f1
- Update CODEOWNERS for layer quantization path (#11818) · 6058fb52
  Lianmin Zheng authored Oct 18, 2025
  
  6058fb52
- Fix: Dynamic RoPE Cache Expansion to Prevent Position-ID Out-of-Bounds in... · 80407b04
  YAMY authored Oct 18, 2025
```
Fix: Dynamic RoPE Cache Expansion to Prevent Position-ID Out-of-Bounds in EAGLE + Long-Sequence Workloads (#10788)
```
  80407b04
- Improve `send_sone` script (#11817) · b288f4f4
  Liangsheng Yin authored Oct 19, 2025
  
  b288f4f4
- fix: do not wrap invalid grammar objects during constrained generation (#11328) · 6d6ea5af
  tazjin authored Oct 19, 2025
  
  6d6ea5af
- make sure logit bias is applied during eagle spec decoding verification (#11555) · 1dacedd2
  Marin authored Oct 19, 2025
  
  1dacedd2
- [1/2][feature] support openai like classification api (#11618) · b5e14b2b
  ybyang authored Oct 19, 2025
  
  b5e14b2b
- [2/2] [feature] support openai like classification api in router (#11670) · d513ee93
  ybyang authored Oct 19, 2025
  
  d513ee93
- [router] Add Configurable L0 and L1 Tokenizer Caching (#11688) · a7ae61ed
  Simo Lin authored Oct 18, 2025
  
  a7ae61ed
18 Oct, 2025 24 commits
- Fix Dockerfile not installing correct version of DeepEP for arm build (#11773) · fda0cb2a
  kyleliang-nv authored Oct 18, 2025
  
  fda0cb2a
- Use cutlass fp4 gemm by default (#11813) · ebda73dc
  Qiaolin Yu authored Oct 18, 2025
  
  ebda73dc
- ci: update `lmms-eval` to speed up multimodal CI (#11000) · f4f8a1b4
  b8zhong authored Oct 18, 2025
  
  f4f8a1b4
- feat(example/fastapi): support --startup-timeout using... · c44e985d
  Kindyaa authored Oct 19, 2025
```
feat(example/fastapi): support --startup-timeout using Qwen3-Next-80B-A3B-Instruct as example (#11710)
Co-authored-by: chenan01 <chenan01@cheche-MacBook-Pro.local>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
```
  c44e985d
- support server arg override KV cache to bf16 to avoid slow cases (#11749) · f9a7d9b3
  b8zhong authored Oct 18, 2025
  
  f9a7d9b3
- [overlap-spec] support page size > 1 (#11772) · a93f10a7
  Liangsheng Yin authored Oct 19, 2025
  
  a93f10a7
- [HiCache] feat: add more eviction policy (#11506) · 585e1223
  Teng Ma authored Oct 18, 2025
  
  585e1223
- Bump torch_memory_saver to avoid installing pre-release versions (#11797) · a7043c6f
  fzyzcjy authored Oct 18, 2025
  
  a7043c6f
- Fix install instructions and pyproject.tomls (#11781) · 67e34c56
  Lianmin Zheng authored Oct 18, 2025
  
  67e34c56
- Eager Compiler for Torch Compile (#11803) · 1d726528
  Yuwei An authored Oct 18, 2025
```
Signed-off-by: Oasis-Git <ayw.sirius19@gmail.com>
```
  1d726528
- set default attention backend for deterministic inference (#11801) · f4488e9d
  Minglei Zhu authored Oct 18, 2025
  
  f4488e9d
- [RL] use cpu group to prepare_mlp_sync_batch_raw when the server is offloaded (#10152) · e68a2b5b
  Zilin Zhu authored Oct 18, 2025
  
  e68a2b5b
- [RL] support weight update with DP attention (#11669) · 31b9f19e
  Zilin Zhu authored Oct 18, 2025
  
  31b9f19e
- fix command line usage of profiling (#11793) · 547003bd
  Qiaolin Yu authored Oct 17, 2025
  
  547003bd
- fix(glm45): disable reduce scatter (#11665) · f7ab9554
  Jimmy authored Oct 18, 2025
```
Co-authored-by: Shangming Cai <csmthu@gmail.com>
```
  f7ab9554
- Try add back no-commit-to-branch (#11799) · dbbd4e18
  fzyzcjy authored Oct 18, 2025
  
  dbbd4e18
- [router][grpc] Support parallel queue puts in grpc_request_manager and remove... · ca240eef
  Chang Su authored Oct 17, 2025
```
[router][grpc] Support parallel queue puts in grpc_request_manager and remove mutex for grpc_client (#11798)
```
  ca240eef
- Enable lint on main (#11794) · 6c7c92eb
  fzyzcjy authored Oct 18, 2025
  
  6c7c92eb
- [Refactor] move `deep_gemm_wrapper` out of `quantization` (#11784) · 5b214b50
  Cheng Wan authored Oct 17, 2025
  
  5b214b50
- completely remove mixed mode deterministic test as prefix mode could cover it (#11783) · 13219e1e
  Minglei Zhu authored Oct 17, 2025
```
Co-authored-by: Baizhou Zhang <sobereddiezhang@gmail.com>
```
  13219e1e
- Make single-batch overlap compatible with offloading (#11614) · 33e9bbec
  fzyzcjy authored Oct 18, 2025
  
  33e9bbec
- Super tiny fix CI (#11788) · dcb8f090
  fzyzcjy authored Oct 18, 2025
  
  dcb8f090
- Set CUDA_VISIBLE_DEVICES to achieve one GPU per process (#9170) · 9eefe2c0
  Lianmin Zheng authored Oct 17, 2025
```
Co-authored-by: SangBin Cho <rkooo567@gmail.com>
Co-authored-by: Cheng Wan <cwan@x.ai>
Co-authored-by: Cheng Wan <54331508+ch-wan@users.noreply.github.com>
```
  9eefe2c0
- Manually flip deepep_mode for cuda_graph (#11666) · 69fe3c97
  Zilin Zhu authored Oct 18, 2025
  
  69fe3c97