Commits · 5c7aa009766983361b2acaf3c3a3864101ab7b8a · change / sglang · GitLab

26 May, 2025 21 commits
- 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
- [FA][Test] Fix Sparse FA test (#6306) · 006ead9d
  Brayden Zhong authored May 26, 2025
  
  006ead9d
- 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
- Revert "Tiny fix lint CI does not trigger on master (#6609)" (#6610) · fc419b62
  Yineng Zhang authored May 25, 2025
  
  fc419b62
- chore: upgrade transformers 4.52.3 (#6575) · 7eb9d8e5
  Yineng Zhang authored May 25, 2025
```
Co-authored-by: Mick <mickjagger19@icloud.com>
```
  7eb9d8e5
- Tiny fix lint CI does not trigger on master (#6609) · 84147254
  fzyzcjy authored May 26, 2025
  
  84147254
- 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
- Tiny change killall_sglang.sh (#6596) · d502dae0
  fzyzcjy authored May 26, 2025
  
  d502dae0
- 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
- Support fake perfectly balanced EP dispatch algorithm (#6571) · 0ca18117
  fzyzcjy authored May 26, 2025
  
  0ca18117
- Fix bench_serving does not support changing warmup requests (#6439) · 2c3a6fe1
  fzyzcjy authored May 26, 2025
  
  2c3a6fe1
- [PD] Fix prefill_servers in mini_lb (#6527) · 8b33d8df
  wangxiyu191 authored May 26, 2025
  
  8b33d8df
25 May, 2025 11 commits
- Fix some issues with current docs. (#6588) · e235be16
  simveit authored May 25, 2025
  
  e235be16
- Hint users when weight update timeouts (#6570) · 5ccf8fe1
  fzyzcjy authored May 26, 2025
  
  5ccf8fe1
- added support for tied weights in qwen pipeline parallelism (#6546) · 3f23d8cd
  Shenggui Li authored May 25, 2025
  
  3f23d8cd
- Sgl-router Prometheus metrics endpoint and usage track metrics (#6537) · 1a399799
  Chao Yang authored May 24, 2025
  
  1a399799
- Support Phi-4 Multi-Modal (text + vision only) (#6494) · 022012aa
  Lifu Huang authored May 24, 2025
  
  022012aa
- [OAI] Support non-normalized logprobs in OpenAI server (#5961) · 681e7af3
  Chang Su authored May 24, 2025
  
  681e7af3
- Refactor vlm embedding routine to use precomputed feature (#6543) · 681fdc26
  Xinyuan Tong authored May 24, 2025
```
Signed-off-by: Xinyuan Tong <justinning0323@outlook.com>
```
  681fdc26
- Support overlapping two batches (#4068) · 0d477880
  fzyzcjy authored May 25, 2025
  
  0d477880
- Utilize static dispatching for communicator (#6577) · f4560373
  fzyzcjy authored May 25, 2025
  
  f4560373
- Add back DeepSeek non-TBO branches (#6578) · b2388433
  fzyzcjy authored May 25, 2025
  
  b2388433
- Refactor attention into multiple stages (#6477) · a38376fa
  fzyzcjy authored May 25, 2025
  
  a38376fa
24 May, 2025 8 commits
- Fix GPU OOM (#6564) · 7a5e6ce1
  kk authored May 25, 2025
```
Co-authored-by: michael <michael.zhang@amd.com>
```
  7a5e6ce1
- Temporarily disable MI325x 8 gpu testing. (#6576) · 24c035f2
  Sai Enduri authored May 24, 2025
  
  24c035f2
- chore: bump v0.4.6.post5 (#6566) · 7e257cd6
  Yineng Zhang authored May 24, 2025
  
  7e257cd6
- Fix accuracy is zero when enabling moe-dense-tp-size as in large scale EP (#6567) · c4831e2f
  fzyzcjy authored May 24, 2025
  
  c4831e2f
- [FIX]remove ServerArgs duplicate code (#6485) · 2e37fa07
  Neo authored May 24, 2025
  
  2e37fa07
- [PD] Support structured output (#6560) · 2d831c6e
  Byron Hsu authored May 23, 2025
  
  2d831c6e
- feat(Tool Calling): Support `required` and specific function mode (#6550) · ed0c3035
  Chang Su authored May 23, 2025
  
  ed0c3035
- support eplb for qwen3 (#6533) · e6f11356
  Yi Zhang authored May 24, 2025
  
  e6f11356