Commits · 60bcbf2a35e290a351386f4dbe85f5b0534acb14 · change / sglang · GitLab

11 Apr, 2025 8 commits
- remove moe_align_block_size torch.zeros in small batch/expert mode (#5298) · 60bcbf2a
  Xiaoyu Zhang authored Apr 12, 2025
  
  60bcbf2a
- [PD] fix: skip warmup request in disaggregation mode to prevent crash on timeout (#5292) · c35dcfdb
  Yusong Gao authored Apr 11, 2025
  
  c35dcfdb
- [fix] fix mrope positions not picked up (#5265) · e53a0b3d
  Mick authored Apr 11, 2025
  
  e53a0b3d
- Support `--enable-llama4-multimodal` (#5254) · 038bc5d5
  Cheng Wan authored Apr 11, 2025
  
  038bc5d5
- Optimize GPU memory usage in FlashAttentionBackend's strided indexing (#5262) · aee62d74
  Chang Su authored Apr 11, 2025
```
Co-authored-by: ch-wan <cwan39@gatech.edu>
```
  aee62d74
- Optimize attention in llama4 (#5127) · cd7e32e2
  fzyzcjy authored Apr 11, 2025
  
  cd7e32e2
- ROCm/AITER CK_MoE: update 2-stage kernels & support both Activations (#5228) · 88799448
  HAI authored Apr 10, 2025
  
  88799448
- Fix torch.compile cacheing (#5259) · a879811c
  Richard Zou authored Apr 10, 2025
```
Co-authored-by: zhyncs <me@zhyncs.com>
```
  a879811c
10 Apr, 2025 6 commits
- Update deps for mllama4 (#5215) · 1078396f
  Ke Bao authored Apr 11, 2025
  
  1078396f
- [PD] Add get_contiguous_buf_infos interface for MLATokenToKVPool (#5204) · 7e4f72dd
  Teng Ma authored Apr 10, 2025
  
  7e4f72dd
- [PD] Support KV transfer with mooncake (#4880) · 4c31ae9f
  Teng Ma authored Apr 10, 2025
```
Signed-off-by: Shangming Cai <caishangming@linux.alibaba.com>
Co-authored-by: Shangming Cai <caishangming@linux.alibaba.com>
Co-authored-by: Xuchun Shang <xuchun.shang@linux.alibaba.com>
Co-authored-by: shangmingc <csmthu@gmail.com>
```
  4c31ae9f
- reduce moe_align_block_size_kernel small batch mode overhead (#5086) · f730362e
  Xiaoyu Zhang authored Apr 10, 2025
  
  f730362e
- Fix DeepSeek error when using DeepEP mode (#5190) · e3c4bd31
  fzyzcjy authored Apr 10, 2025
  
  e3c4bd31
- [metrics] Add in queue metrics (#4444) · 5db37c86
  Stefan He authored Apr 09, 2025
  
  5db37c86
09 Apr, 2025 15 commits
- fix: log warning when disable cuda graph (#5209) · 4cb53ecd
  Yineng Zhang authored Apr 09, 2025
  
  4cb53ecd
- Add H20 dtype fp8_w8a8 fused MoE kernel tuning configs for DeepSeek V3/R1 (#5196) · 456b008b
  Zhaoyang Hao authored Apr 10, 2025
  
  456b008b
- update grok test (#5171) · 7f875f12
  saienduri authored Apr 09, 2025
  
  7f875f12
- model: support mllama4 (#5144) · fbebcb7a
  Mick authored Apr 10, 2025
  
  fbebcb7a
- Support Llama4 fp8 inference (#5194) · 40652482
  HandH1998 authored Apr 09, 2025
```
Co-authored-by: laixinn <xielx@shanghaitech.edu.cn>
Co-authored-by: sleepcoo <sleepcoo@gmail.com>
Co-authored-by: zhyncs <me@zhyncs.com>
```
  40652482
- Optimize topk operation in llama4 (#5128) · 86a876d8
  fzyzcjy authored Apr 09, 2025
  
  86a876d8
- Fix ci test "test_eval_fp8_accuracy" failed (#5185) · 92823069
  kk authored Apr 09, 2025
```
Co-authored-by: wunhuang <wunhuang@amd.com>
```
  92823069
- Let `bench_one_batch` support `enable_dp_attention` (#4058) · 61970b08
  fzyzcjy authored Apr 09, 2025
  
  61970b08
- [DeepEP] fix: import buffer error (#5179) · 76c48a09
  Cheng Wan authored Apr 09, 2025
  
  76c48a09
- fix: use DeepEPDispatcher on CUDA (#5180) · 90caf06c
  Yineng Zhang authored Apr 08, 2025
  
  90caf06c
- feat: add DeepGEMM build warning (#5176) · 6669d127
  Yineng Zhang authored Apr 08, 2025
```
Co-authored-by: grimoire <streetyao@live.com>
```
  6669d127
- [Fix] DeepEP Compatibility with Low Latency (#5068) · bc3f6db2
  Jinyan Chen authored Apr 09, 2025
```
Co-authored-by: ch-wan <cwan39@gatech.edu>
```
  bc3f6db2
- [Bugfix] Fix index out of bounds in local attention with large sequences (#5173) · aac531c5
  Chang Su authored Apr 08, 2025
  
  aac531c5
- Fix multimodal hashing error (#5174) · 466899e6
  fzyzcjy authored Apr 09, 2025
  
  466899e6
- FP4 weight loading and inference (2/2) (#3972) · 11d760d5
  Trevor Morris authored Apr 08, 2025
  
  11d760d5
08 Apr, 2025 11 commits
- Support 2x8xH100 for Llama 4 (#5159) · 5039d547
  fzyzcjy authored Apr 09, 2025
  
  5039d547
- [feat&refactor] Enhance multimodal input support with refactor io_struct (#4938) · d09a51f1
  XinyuanTong authored Apr 08, 2025
```
Signed-off-by: Xinyuan Tong <justinning0323@outlook.com>
```
  d09a51f1
- [PD] Simplify mini LB (#4911) · 6d3b35fa
  Byron Hsu authored Apr 08, 2025
```
Co-authored-by: Liangsheng Yin <hnyls2002@gmail.com>
```
  6d3b35fa
- [PD] Fix unclosed prefill connection warning of mini_lb (#5155) · 89a55418
  shangmingc authored Apr 09, 2025
```
Signed-off-by: Shangming Cai <caishangming@linux.alibaba.com>
```
  89a55418
- Fix loading KV quantization scale; Enable modelopt kv cache (#4686) · 2695ab05
  Yun Dai authored Apr 08, 2025
```
Co-authored-by: qingquansong <ustcsqq@gmail.com>
```
  2695ab05
- Fix torch compile errors (#5158) · 88d6fd9a
  kk authored Apr 08, 2025
  
  88d6fd9a
- fix empty_cache error in pt_weights_iterator (#5151) · cc88d98a
  DangKai authored Apr 08, 2025
```
Co-authored-by: dangkai.dk <dangkai.dk@alibaba-inc.com>
```
  cc88d98a
- Add unit test on page_size > 1 and mla and integration test for Flash Attention 3 (#4760) · 804d9f2e
  Yubo Wang authored Apr 07, 2025
  
  804d9f2e
- [FA3 Feature] Support multi modal Llama-3.2-11B-Vision-Instruct (#5103) · a7c3f74b
  Chunan Zeng authored Apr 07, 2025
  
  a7c3f74b
- Fix run time error in ROCm platform (#5147) · 5a144a8a
  kk authored Apr 08, 2025
```
Co-authored-by: wunhuang <wunhuang@amd.com>
Co-authored-by: root <root@dell300x-pla-t10-17.pla.dcgpu>
```
  5a144a8a
- fix multimodal hash feature (#5083) · 27f8e6b9
  huangtingwei authored Apr 08, 2025
  
  27f8e6b9