Commits · 7ddf8e83d2d9f1d423bc3cae999ef9366fd0959c · change / sglang · GitLab

16 Jun, 2025 9 commits
- [EAGLE] Fix draft kv cache layout for fa3 and topk > 1 (#7239) · 7ddf8e83
  Lianmin Zheng authored Jun 16, 2025
  
  7ddf8e83
- Release sgl-kernel 0.1.9 (#7232) · 8321f8e4
  Lianmin Zheng authored Jun 16, 2025
  
  8321f8e4
- Fix sampling for speculative decoding & simplify kernels (#7207) · cfceb83d
  Lianmin Zheng authored Jun 16, 2025
  
  cfceb83d
- [EAGLE] Refactor code for page size > 1 & more simplifications (#7213) · b1286a11
  Lianmin Zheng authored Jun 16, 2025
  
  b1286a11
- Minor style and doc fix (#7228) · 21615cc3
  Lianmin Zheng authored Jun 16, 2025
  
  21615cc3
- refine fused_moe benchmark (#7221) · 0ae1e9a7
  Xiaoyu Zhang authored Jun 16, 2025
  
  0ae1e9a7
- Support LoRA in MMMU benchmark script. (#7218) · e07d0647
  Lifu Huang authored Jun 15, 2025
  
  e07d0647
- Implement gather before attn (#6378) · 3c2274fb
  Cheng Wan authored Jun 15, 2025
  
  3c2274fb
- Fix ChunkCache object has no attribute 'disable' (#7217) · d2679f51
  Baizhou Zhang authored Jun 15, 2025
  
  d2679f51
15 Jun, 2025 14 commits
- Minor PD style fix (#7215) · 96be97bf
  Byron Hsu authored Jun 15, 2025
  
  96be97bf
- [PD] use int32 for kv indices & get num_reserved_decode_tokens from server_args (#7214) · 88f9c347
  Byron Hsu authored Jun 15, 2025
  
  88f9c347
- Revert "[EAGLE] Refactor code for page size > 1 & more simplifications" (#7210) · fff10809
  Lianmin Zheng authored Jun 15, 2025
  
  fff10809
- [EAGLE] Refactor code for page size > 1 & more simplifications (#7163) · 5f1ab327
  Lianmin Zheng authored Jun 14, 2025
  
  5f1ab327
- feat: use zstd for docker (#7205) · 7df7c679
  Yineng Zhang authored Jun 14, 2025
  
  7df7c679
- Fix grammar abort & Minor style fixes (#7204) · 38af4f68
  Lianmin Zheng authored Jun 14, 2025
  
  38af4f68
- Lianmin/simplify memory pool (#7202) · a6305c7d
  Lianmin Zheng authored Jun 14, 2025
  
  a6305c7d
- Move host memory pools into a separate file (#7200) · a023856b
  Lianmin Zheng authored Jun 14, 2025
  
  a023856b
- [PD] Support decode retract and update decode.py (#7196) · db0cc57e
  Byron Hsu authored Jun 14, 2025
  
  db0cc57e
- Fix error when disabling new DeepGEMM (#7198) · 349bb2c9
  fzyzcjy authored Jun 15, 2025
  
  349bb2c9
- Fix NCCL 2.27.3 not in docker image (#7195) · 0b8939bc
  fzyzcjy authored Jun 15, 2025
  
  0b8939bc
- chore: upgrade sgl-kernel v0.1.8.post2 (#7186) · ed89837c
  JieXin Liang authored Jun 15, 2025
```
Co-authored-by: zhyncs <me@zhyncs.com>
```
  ed89837c
- [fix] fix determine_num_fused_shared_experts (#7180) · 55561e25
  JieXin Liang authored Jun 15, 2025
  
  55561e25
- chore: bump v0.1.8.post2 (#7189) · 44733203
  Yineng Zhang authored Jun 14, 2025
  
  44733203
14 Jun, 2025 13 commits
- Fix a minor bug related to DeepGEMM upgrade (#7191) · 0bd67ba2
  Zhijian Liu authored Jun 14, 2025
  
  0bd67ba2
- [PD] Update prefill.py (#7190) · 7d316991
  Byron Hsu authored Jun 14, 2025
  
  7d316991
- [fix] fix cutlass_mla_backend with cuda_graph and add sm_scale for sgl-kernel cutlass_mla (#7184) · ab1a4fa5
  JieXin Liang authored Jun 15, 2025
  
  ab1a4fa5
- [fix] fix dsv3 weight loader tqdm and simplify shared experts fusion (#7181) · ed54bf9d
  JieXin Liang authored Jun 15, 2025
  
  ed54bf9d
- Fix shared experts fusion + weight requant (#7177) · b57d87c2
  fzyzcjy authored Jun 14, 2025
  
  b57d87c2
- Add Phi-4-mm to supported VLM supported model list. (#7178) · 98538822
  Lifu Huang authored Jun 13, 2025
  
  98538822
- Increase timeout in test/srt/test_disaggregation.py (#7175) · f47a1b1d
  Lianmin Zheng authored Jun 13, 2025
  
  f47a1b1d
- Support new DeepGEMM (#7172) · 93cec433
  fzyzcjy authored Jun 14, 2025
  
  93cec433
- Improve test cases for eagle infer (#7173) · ba589b88
  Lianmin Zheng authored Jun 13, 2025
  
  ba589b88
- Add test for refactored openai server (#7161) · 50876abc
  Jinn authored Jun 13, 2025
  
  50876abc
- Refactor DeepGEMM integration (#7150) · b4c41f72
  fzyzcjy authored Jun 14, 2025
  
  b4c41f72
- Support new DeepGEMM input format in silu_and_mul_masked_post_quant_fwd (#7153) · 8b8f2e74
  fzyzcjy authored Jun 14, 2025
  
  8b8f2e74
- Split the eagle test into two files (#7170) · 0fc3d992
  Lianmin Zheng authored Jun 13, 2025
  
  0fc3d992
13 Jun, 2025 4 commits
- Minor style change of triton backend (#7165) · be2d985d
  Lianmin Zheng authored Jun 13, 2025
  
  be2d985d
- Re-quantize DeepSeek model weights to support DeepGEMM new input format (#7156) · 5b1afa78
  fzyzcjy authored Jun 14, 2025
  
  5b1afa78
- Remove 200us slow concat kernel (part 2: srt) (#7020) · c49c1d92
  fzyzcjy authored Jun 14, 2025
  
  c49c1d92
- Tiny add sanity checks for DeepGEMM inputs (#7157) · 0f1dfa1e
  fzyzcjy authored Jun 14, 2025
  
  0f1dfa1e