Commits · c4831e2fcfcf76215f5645029fc64b20f613a002 · change / sglang

"git@developer.sourcefind.cn:change/sglang.git" did not exist on "d5ae2ebaa2a1021b53caa25aa541ca39d590918d"

24 May, 2025 6 commits
- 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 13 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
- [RL] Remove the w13 weight_scale and input_scale for UnquantizedEPMoE… (#6308) · e9feb488
  Zilin Zhu authored May 22, 2025
  
  e9feb488
- Support updating expert locations dynamically (#6388) · fc992a09
  fzyzcjy authored May 22, 2025
  
  fc992a09
- [PD] Abort request if transfer fails (#6504) · 3bde1010
  Byron Hsu authored May 21, 2025
  
  3bde1010
- [PD] Add doc and simplify sender.send (#6019) · 75135580
  Byron Hsu authored May 21, 2025
  
  75135580
- Apply constraint grammar to EAGLE (#6499) · 6ce0ed07
  Ke Bao authored May 22, 2025
```
Co-authored-by: merrymercy <lianminzheng@gmail.com>
```
  6ce0ed07
- Recover from corrupted cache file in bench serving (#6510) · 969660c7
  fzyzcjy authored May 22, 2025
  
  969660c7
21 May, 2025 13 commits
- fix: EXAONE when using tie_word_embeddings (#5759) · ada268fd
  Kyungmin Lee authored May 22, 2025
  
  ada268fd
- [Fix]Fix capture fail bug for DeepSeek (#6275) · d4c038da
  Baizhou Zhang authored May 21, 2025
  
  d4c038da
- Fix bench_one_batch_server (#6503) · 55f6005f
  fzyzcjy authored May 22, 2025
  
  55f6005f
- Let bench_one_batch_server use sharegpt data to make expert distribution more natural (#5573) · 7222e1da
  fzyzcjy authored May 21, 2025
  
  7222e1da
- Support redundant experts in expert parallel (#6461) · ccfe5c00
  fzyzcjy authored May 21, 2025
  
  ccfe5c00
- Tiny add stage assertions to DeepEPDispatcher to avoid misuse (#6467) · a071dc40
  fzyzcjy authored May 21, 2025
  
  a071dc40
- Fix num_qps_per_rank computation when providing custom DeepEP configuration (#6468) · a40aecc5
  fzyzcjy authored May 21, 2025
  
  a40aecc5
- Refactor DeepSeek attention dispatching (#6476) · d6e1d28c
  fzyzcjy authored May 21, 2025
  
  d6e1d28c
- [RL] allow weight updation with dp attention enabled (#6311) · 7c347259
  Zilin Zhu authored May 21, 2025
  
  7c347259
- Implement Siglip Vision model, and support BNB quantization for gemma3-mm (#5339) · 4024e1d2
  Jiajun Li authored May 20, 2025
  
  4024e1d2
- aiter attention-backend (default enabled on AMD/ROCm) (#6381) · 5c0b38f3
  HAI authored May 20, 2025
  
  5c0b38f3
- Refactor group_concurrent_contiguous in NIXL (#6214) · 30ca18f4
  Yuan Luo authored May 21, 2025
```
Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>
```
  30ca18f4
- Disable all two stream overlap on amd (#6475) · 03886917
  Lianmin Zheng authored May 20, 2025
  
  03886917
20 May, 2025 1 commit
- Fix master CI for DeepSeek (#6447) · 13feffd0
  fzyzcjy authored May 20, 2025
  
  13feffd0