Commits · 5b5c7237c8ffdb30f08bf11258126c08189ab8e1 · change / sglang · GitLab

16 Apr, 2025 6 commits
- chore: bump v0.4.5.post1 (#5445) · 5b5c7237
  Yineng Zhang authored Apr 15, 2025
  
  5b5c7237
- Support MHA with chunked prefix cache for DeepSeek chunked prefill (#5113) · a42736bb
  Baizhou Zhang authored Apr 15, 2025
  
  a42736bb
- [Bug fix] need record start time in pd mode (#5425) · dd83e7e9
  ybyang authored Apr 16, 2025
  
  dd83e7e9
- [Minor] Move torch.compile patch to a better place (#5397) · 0769b14b
  Lianmin Zheng authored Apr 15, 2025
  
  0769b14b
- Support BNB quantization for llama/mllama (#5038) · bc24205b
  ryang authored Apr 16, 2025
```
Co-authored-by: Yuhao Yang <yyh073@foxmail.com>
```
  bc24205b
- Fix ignore_eos parameter when loading a chat template (#5264) · 27a009bb
  Chang Su authored Apr 15, 2025
  
  27a009bb
15 Apr, 2025 15 commits
- chore: upgrade sgl-kernel 0.0.9.post1 (#5436) · 8ec0bb7d
  Yineng Zhang authored Apr 15, 2025
  
  8ec0bb7d
- feat: update model_specific_adjustment (#5344) · fa909dc3
  Yineng Zhang authored Apr 15, 2025
```
Co-authored-by: hebiao064 <hebiaobuaa@gmail.com>
```
  fa909dc3
- [PD] Remove unused bootstrap param and fix port table type (#5423) · f1b3b75f
  shangmingc authored Apr 15, 2025
  
  f1b3b75f
- Distinguish bootstrap key only in decode server (#5422) · 33b16ad1
  Liangsheng Yin authored Apr 15, 2025
  
  33b16ad1
- [PD] Fix dynamic port support and MLA buffer for Mooncake (#5415) · ffde65a0
  shangmingc authored Apr 15, 2025
```
Signed-off-by: Shangming Cai <caishangming@linux.alibaba.com>
Co-authored-by: ybyang <ybyang7@iflytek.com>
```
  ffde65a0
- Fix broadcast use cuda device lead to memory capacity unbalanced (#5416) · 471650de
  lambert0312 authored Apr 15, 2025
  
  471650de
- Support dynamic connection and TP 16 (#5351) · d06a83fb
  Yuan Luo authored Apr 15, 2025
```
Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>
```
  d06a83fb
- [FIX] Fix concatenation error in capture_bs when open... · 5d134401
  Zhaoyang Hao authored Apr 15, 2025
```
[FIX] Fix concatenation error in capture_bs when open --disable-cuda-graph-padding and without MTP (#5412)
```
  5d134401
- Fix bench_serving with random-ids (#5214) · 3dfc6023
  Yuhong Guo authored Apr 15, 2025
  
  3dfc6023
- Tiny fix DeepseekScalingRotaryEmbedding always use forward_native (#5406) · 15e91d72
  fzyzcjy authored Apr 15, 2025
  
  15e91d72
- chore: upgrade sgl-kernel 0.0.9 (#5401) · 8aab7fdb
  Yineng Zhang authored Apr 14, 2025
  
  8aab7fdb
- [fix/misc] remove duplicate row in deepseek v2 model (#5279) · ee9d6ca6
  Yangcheng Li authored Apr 15, 2025
  
  ee9d6ca6
- Add H20 dtype fp8_w8a8 shared experts fused MoE kernel tuning configs for DeepSeek V3/R1 (#5291) · 2dd64894
  Ximingwang-09 authored Apr 15, 2025
```
Co-authored-by: ximing.wxm <ximing.wxm@antgroup.com>
```
  2dd64894
- Add A800 shared experts fused MoE kernel tuning configs for DeepSeek V3/R1 (#5368) · 61e7c4dd
  lambert0312 authored Apr 15, 2025
  
  61e7c4dd
- [Fix] Turn off DeepGEMM by default (#5263) · f6772f14
  Baizhou Zhang authored Apr 14, 2025
  
  f6772f14
14 Apr, 2025 8 commits
- apply fused moe gate in ds v3/r1 (#5371) · 38076dea
  Xiaoyu Zhang authored Apr 15, 2025
```
Co-authored-by: Yineng Zhang <me@zhyncs.com>
```
  38076dea
- Apply deepseek cuda rope (#5385) · 5e0a9b09
  Ke Bao authored Apr 15, 2025
```
Co-authored-by: Yineng Zhang <me@zhyncs.com>
```
  5e0a9b09
- [perf] experimental enhance fp8 per-tensor quant (#5370) · bdde2375
  JieXin Liang authored Apr 15, 2025
  
  bdde2375
- [PD Bug] fix MLA get_contiguous_buf_infos error (#5384) · e9fc2ac7
  ybyang authored Apr 14, 2025
  
  e9fc2ac7
- Fix PD disaggregation bugs (#5326) · 44afde82
  Liangsheng Yin authored Apr 14, 2025
  
  44afde82
- Support for Qwen2.5-VL Model in bitsandbytes Format (#5003) · 072df753
  yhyang201 authored Apr 14, 2025
  
  072df753
- Fix DeepSeek DP Attention + torch compile (#5367) · defede50
  fzyzcjy authored Apr 14, 2025
```
Co-authored-by: ispobock <ispobaoke@163.com>
```
  defede50
- Free metadata_buffer_index after transfer finished (#5364) · 14e8bd88
  Yongtong Wu authored Apr 14, 2025
  
  14e8bd88
13 Apr, 2025 8 commits
- [DeepEP] Reduce routed scaling overhead (#5277) · adca585b
  yulei authored Apr 14, 2025
```
Co-authored-by: Cheng Wan <54331508+ch-wan@users.noreply.github.com>
```
  adca585b
- fix #5322 (#5359) · 39e41138
  Yineng Zhang authored Apr 13, 2025
  
  39e41138
- fix MLATokenToKVPoolHost get_size_per_token bug (#5161) · 5fbafbb8
  huangtingwei authored Apr 14, 2025
```
Co-authored-by: AniZpZ <zhuangsen.zp@antgroup.com>
```
  5fbafbb8
- [PD] Add transfer backend abstraction (#5328) · a9499885
  Byron Hsu authored Apr 13, 2025
  
  a9499885
- Fix typo: infight -> inflight (#5357) · f7655790
  Liangsheng Yin authored Apr 14, 2025
  
  f7655790
- chore: upgrade sgl-kernel 0.0.8.post3 (#5342) · f58b929a
  Yineng Zhang authored Apr 13, 2025
  
  f58b929a
- Fix: Ensure tensors for dist.broadcast match NCCL backend device (#5322) · 8311b07f
  mlmz authored Apr 13, 2025
  
  8311b07f
- fix: determine if flashinfer is installed (#5336) · 7d3b7c87
  Yineng Zhang authored Apr 12, 2025
  
  7d3b7c87
12 Apr, 2025 3 commits
- Support server based rollout in Verlengine (#4848) · bc92107b
  tianlian yi authored Apr 13, 2025
```
Co-authored-by: Jin Pan <jpan236@wisc.edu>
Co-authored-by: Chayenne <zhaochen20@outlook.com>
Co-authored-by: Jinn <47354855+jhinpan@users.noreply.github.com>
```
  bc92107b
- refine fused_moe tuning docs (#5294) · 3e4794aa
  Xiaoyu Zhang authored Apr 13, 2025
  
  3e4794aa
- Delete python/sglang/srt/layers/moe/fused_moe_triton/configs/E=257,N=… (#5321) · 690ec205
  Xiaoyu Zhang authored Apr 13, 2025
  
  690ec205