Commits · 7eb9d8e5947b4232daf3ce708cbd728e2c677999 · change / sglang · GitLab

26 May, 2025 9 commits
- 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
- 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 9 commits
- 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
- 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
- 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
- [Bugfix](gemma3_mm): handle flatten_batch constraint for multiple images (#6562) · 7b02c326
  Chang Su authored May 23, 2025
  
  7b02c326
23 May, 2025 7 commits
- Update cmdline --enable-dp-attention help string for Qwen 2/3 Moe models. (#6524) · fefa19fe
  miter authored May 24, 2025
```
Signed-off-by: miter <miterv@outlook.com>
```
  fefa19fe
- [PD] Support logprob & Add failure test (#6558) · 8233cc10
  Byron Hsu authored May 23, 2025
  
  8233cc10
- [2/2] Support Qserve (#6521) · 1b2e8f76
  HandH1998 authored May 24, 2025
  
  1b2e8f76
- [PD] support spec decode (#6507) · d2e0881a
  Byron Hsu authored May 23, 2025
```
Co-authored-by: SangBin Cho <rkooo567@gmail.com>
```
  d2e0881a
- Fix topk inference performance reduce (#6474) · 2f427491
  Li Hui authored May 23, 2025
  
  2f427491
- [VLM] Support chunk prefill for VLM (#6355) · 4685fbb8
  Chang Su authored May 22, 2025
```
Co-authored-by: yizhang2077 <1109276519@qq.com>
```
  4685fbb8
- [PD] Fix failure abort (#6535) · 0a4fc73b
  Byron Hsu authored May 22, 2025
  
  0a4fc73b
22 May, 2025 7 commits
- misc: fix accept_length (#6536) · a6970a17
  Yineng Zhang authored May 22, 2025
  
  a6970a17
- Support XiaomiMiMo inference with mtp (#6059) · a6ae3af1
  ryang authored May 23, 2025
  
  a6ae3af1
- chore: upgrade sgl-kernel v0.1.4 (#6532) · 0b07c4a9
  Yineng Zhang authored May 22, 2025
  
  0b07c4a9
- Support qwen3 deepep (#6120) · fc0e3b91
  lukec authored May 23, 2025
  
  fc0e3b91
- Fix missing http status import for PD failure handler (#6520) · 58f10679
  shangmingc authored May 22, 2025
```
Signed-off-by: Shangming Cai <caishangming@linux.alibaba.com>
```
  58f10679
- Support dynamically rebalancing experts using EPLB (#6469) · 7a80f565
  fzyzcjy authored May 22, 2025
  
  7a80f565
- Support logging expert balancedness metrics (#6482) · 9484eba4
  fzyzcjy authored May 22, 2025
  
  9484eba4