Commits · c087ddd6865a52634326a05af66429cb5531cd16 · change / sglang · GitLab

28 May, 2025 1 commit
- Refine pre_reorder_triton_kernel slightly to improve performance (#6627) · c087ddd6
  Yuan Luo authored May 28, 2025
```
Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>
```
  c087ddd6
27 May, 2025 24 commits
- feat: Add warnings for invalid tool_choice and UTs (#6582) · 41ba767f
  Chang Su authored May 27, 2025
  
  41ba767f
- fix(tool call): Fix tool_index in PythonicDetector and issues with mixed... · bdb962d7
  Chang Su authored May 27, 2025
```
fix(tool call): Fix tool_index in PythonicDetector and issues with mixed output in non-streaming (#6678)
```
  bdb962d7
- Support gathering expert distribution details (#6665) · 87068b5c
  fzyzcjy authored May 28, 2025
  
  87068b5c
- Fix DeepEP error in Qwen 3 MoE models (#6673) · a564e001
  fzyzcjy authored May 28, 2025
  
  a564e001
- [PD] Make bootstrap code common between NIXL and Mooncake (#6473) · e806f708
  Trevor Morris authored May 27, 2025
  
  e806f708
- Revert "fix communicator for non-dp lm head (#6662)" (#6677) · fa6723f0
  Yineng Zhang authored May 27, 2025
  
  fa6723f0
- Speed up expert location update (#6661) · 673ff668
  fzyzcjy authored May 28, 2025
  
  673ff668
- Fix OOM when updating expert locations (#6660) · 447be242
  fzyzcjy authored May 28, 2025
  
  447be242
- DeepSeek: enable none block-quant FP8 quantizations (#6638) · 183d9f96
  HAI authored May 27, 2025
  
  183d9f96
- Support EAGLE draft extend CUDA graph (#6606) · 63195028
  Ke Bao authored May 27, 2025
```
Co-authored-by: Sehoon Kim <sehoonkim@berkeley.edu>
```
  63195028
- fix communicator for non-dp lm head (#6662) · a3d7f4b6
  Cheng Wan authored May 27, 2025
  
  a3d7f4b6
- Fix qwen3 tbo/dp-lm-head (#6652) · b18416fb
  Yi Zhang authored May 27, 2025
  
  b18416fb
- fix: fix nightly test from updating transformers (#6658) · ce9d690e
  Mick authored May 27, 2025
  
  ce9d690e
- Add environment flag for disabling message queue broadcaster (#6403) · bdaefbbf
  Baizhou Zhang authored May 26, 2025
  
  bdaefbbf
- [Bugfix]: Fix call for function_call_parser.multi_format_detector in adapter.py (#6650) · ae335842
  Chang Su authored May 26, 2025
  
  ae335842
- Refactor LoRA handling to support adapter tensors in fused format (#6585) · 477a101c
  Lifu Huang authored May 26, 2025
  
  477a101c
- Super tiny rename environment variable (#6648) · 1a8f5f68
  fzyzcjy authored May 27, 2025
  
  1a8f5f68
- Support TP in attention for two batch overlap (#6634) · 32cd7070
  fzyzcjy authored May 27, 2025
  
  32cd7070
- Tiny refactor communicator (#6646) · ebd1ed49
  fzyzcjy authored May 27, 2025
  
  ebd1ed49
- [New Model] Devstral support (#6547) · d6864ce6
  Xinyuan Tong authored May 26, 2025
```
Signed-off-by: Xinyuan Tong <justinning0323@outlook.com>
```
  d6864ce6
- fix: added "\n" to qwen25 tool parser structural tags (#6631) · 755a3661
  Shi Shuai authored May 27, 2025
  
  755a3661
- follow-up: move Idefics2 to a shared location to eliminate unexpected dependency. (#6603) · 79a39ac0
  Lifu Huang authored May 26, 2025
  
  79a39ac0
- [PD] Handle P/D failure and reconnect without affecting other instances (#6263) · 3ce94f71
  shangmingc authored May 27, 2025
```
Signed-off-by: Shangming Cai <caishangming@linux.alibaba.com>
```
  3ce94f71
- Tiny fix sampler error when prob is not contiguous (#6639) · ca95556c
  fzyzcjy authored May 27, 2025
  
  ca95556c
26 May, 2025 15 commits
- Tiny fix missing expert location dispatch info (#6620) · 0ca3e568
  fzyzcjy authored May 26, 2025
  
  0ca3e568
- Fix EPLB algorithm fail to run when using 3 nodes for prefill (#6629) · 5c7aa009
  fzyzcjy authored May 26, 2025
  
  5c7aa009
- Automatically configure for EPLB-related args (#6628) · fe386aca
  fzyzcjy authored May 26, 2025
  
  fe386aca
- fix qwen3moe eplb prefill bug (#6617) · 14d1075f
  Yi Zhang authored May 26, 2025
  
  14d1075f
- Supported precomputed feature for Kimi VL (#6599) · 0d503090
  Lifu Huang authored May 26, 2025
  
  0d503090
- Tiny fix CI (#6611) · 501efc3d
  fzyzcjy authored May 26, 2025
  
  501efc3d
- qwen3moe support two batch overlap (#6598) · f9bab3d5
  Yi Zhang authored May 26, 2025
  
  f9bab3d5
- feat: Improve Mistral and Qwen25 function call parsing (#6597) · 16f69b1f
  Chang Su authored May 25, 2025
  
  16f69b1f
- refactor qwen moe code, use communicator to support tp+dp (#6581) · 65f09131
  Yi Zhang authored May 26, 2025
  
  65f09131
- chore: upgrade transformers 4.52.3 (#6575) · 7eb9d8e5
  Yineng Zhang authored May 25, 2025
```
Co-authored-by: Mick <mickjagger19@icloud.com>
```
  7eb9d8e5
- Support accurate length control for bench serving (#6594) · 6bebef60
  fzyzcjy authored May 26, 2025
  
  6bebef60
- Auto handle PD disaggregation in bench_serving (#6587) · 25be63d0
  fzyzcjy authored May 26, 2025
```
Co-authored-by: yizhang2077 <1109276519@qq.com>
```
  25be63d0
- Logging and minor fixes to two batch overlap and EPLB (#6595) · 93e53f6e
  fzyzcjy authored May 26, 2025
  
  93e53f6e
- Improve performance of two batch overlap in some imbalanced cases (#6593) · a191a0e4
  fzyzcjy authored May 26, 2025
  
  a191a0e4
- Fix profiling will crash the server when using num_steps (#6586) · 8c7279c2
  fzyzcjy authored May 26, 2025
  
  8c7279c2