Commits · a1c1ebe9357418ae8cf8ffdf66e3eaec066170e8 · change / sglang · GitLab

25 Jun, 2025 19 commits
- Fix FP8 KV Cache Support in FA3 Backend (#7148) · a1c1ebe9
  Yuhong Guo authored Jun 25, 2025
  
  a1c1ebe9
- minor: 'role' must be system/assistant/tool, but case insensitive for now (#7499) · fe2a0f96
  mlmz authored Jun 25, 2025
  
  fe2a0f96
- feat: add return hidden_states at async generation (#7507) · 20beb370
  eigen authored Jun 25, 2025
  
  20beb370
- Fix typo of flash_cache (#7513) · 00fbd8a4
  Stefan He authored Jun 25, 2025
  
  00fbd8a4
- take aiter get_rope back (#7521) · 802815e4
  valarLip authored Jun 25, 2025
  
  802815e4
- enable aiter fp8 blockscale quant (#7520) · 4c6675c4
  valarLip authored Jun 25, 2025
  
  4c6675c4
- [PD] Add different TP sizes support for no-MLA models (#6793) · e21aa1df
  Hongbo Xu authored Jun 25, 2025
```
Co-authored-by: shangmingc <csmthu@gmail.com>
Co-authored-by: Shangming Cai <caishangming@linux.alibaba.com>
```
  e21aa1df
- feat: send kvmetrics from sglang scheduler (#6721) · f3cbd245
  zixuanzhang226 authored Jun 25, 2025
  
  f3cbd245
- npu fused op (#7386) · 506a2d59
  ll819214 authored Jun 25, 2025
```
Co-authored-by: Li Junwen <lijunwen13@hisilicon.com>
```
  506a2d59
- [CI] Upgrade mooncake to v0.3.4.post2 to fix potential slice failed bug (#7522) · a07f8ae4
  Shangming Cai authored Jun 25, 2025
```
Signed-off-by: Shangming Cai <caishangming@linux.alibaba.com>
```
  a07f8ae4
- [CPU] [BF16] Call fused_experts_cpu, weight_packed_linear and bmm_cpu kernel... · 7eb47b0f
  Chunyuan WU authored Jun 25, 2025
```
[CPU] [BF16] Call fused_experts_cpu, weight_packed_linear and bmm_cpu kernel in DeepSeek model (#6641)
Co-authored-by: Thien Tran <gau.nernst@yahoo.com.sg>
```
  7eb47b0f
- fix: force synchronization between TP workers when update_weights (#6626) · bc2e5645
  DangKai authored Jun 25, 2025
```
Co-authored-by: dangkai.dk <dangkai.dk@alibaba-inc.com>
```
  bc2e5645
- [ci] add router benchmark script and CI (#7498) · 3abc3036
  Simo Lin authored Jun 25, 2025
  
  3abc3036
- clean duplicate code (#7512) · afeed465
  linzhuo authored Jun 25, 2025
  
  afeed465
- EPLB support for MTP (#7510) · 587b4c6e
  yilian49 authored Jun 25, 2025
  
  587b4c6e
- [PD][Spec] Fix hidden state transfer for spec decode (#7516) · 7b9a174a
  Shangming Cai authored Jun 25, 2025
```
Signed-off-by: Shangming Cai <caishangming@linux.alibaba.com>
```
  7b9a174a
- [OAI] patch origin request_id logic (#7508) · 03c039c4
  ybyang authored Jun 25, 2025
  
  03c039c4
- Fuse sorted_token_ids padding to moe_align_block_size kernel (#7437) · 57ab7769
  Ke Bao authored Jun 25, 2025
  
  57ab7769
- misc: Improvement to serving_chat.py and add more ut (#7489) · 112b496a
  Chang Su authored Jun 24, 2025
  
  112b496a
24 Jun, 2025 14 commits
- fix: Add `--model` as an alias for `--model-path` in server_args (#7505) · 3562256b
  Chang Su authored Jun 24, 2025
  
  3562256b
- [PD] NIXL: Register kv args in advance and cleanup finished requests (#6717) · 5f527834
  Trevor Morris authored Jun 24, 2025
  
  5f527834
- Support multi-thread model weight loading (#7277) · 9f1787fa
  xianzhiT authored Jun 25, 2025
  
  9f1787fa
- [benchmark] fbgemm benchmark support bandwidth report and support fbgemm_cutlass_gmm (#7422) · 8ecad0b1
  Xiaoyu Zhang authored Jun 25, 2025
  
  8ecad0b1
- Remove cumsum_buffer initilization (#7439) · 7151194b
  Ke Bao authored Jun 24, 2025
  
  7151194b
- [PD Disaggregation] replace transfer with batch transfer for better performance (#7236) · 2ed68d7a
  Francis authored Jun 24, 2025
  
  2ed68d7a
- enable aiter_biased_grouped_topk kernel (#7423) · e984d507
  valarLip authored Jun 24, 2025
  
  e984d507
- [AMD] add aiter fused moe in DeepEP path (#7268) · 755f3147
  Alex Sun authored Jun 24, 2025
  
  755f3147
- chore: bump v0.4.8 (#7493) · 7c3a12c0
  Yineng Zhang authored Jun 23, 2025
  
  7c3a12c0
- chore: bump sgl-kernel v0.2.0 (#7490) · e846d95e
  Yineng Zhang authored Jun 23, 2025
  
  e846d95e
- Fix MTP with Deepseek R1 Fp4 (#7376) · 15f34013
  Charles Chen authored Jun 23, 2025
  
  15f34013
- Fix RequestValidationError response format (#7487) · d04163b3
  Chang Su authored Jun 23, 2025
  
  d04163b3
- bugfix: Prevent global mutation of conv.stop_str across requests (#7347) · 7732bbe4
  huangtingwei authored Jun 24, 2025
```
Co-authored-by: Chang Su <chang.s.su@oracle.com>
```
  7732bbe4
- Perormance: Enable cuda graph for dp idle batch (#7269) · ed0a0b69
  u4lr451 authored Jun 24, 2025
```
Co-authored-by: austindeng <austindeng@tencent.com>
Co-authored-by: Cheng Wan <54331508+ch-wan@users.noreply.github.com>
Co-authored-by: ch-wan <cwan39@gatech.edu>
```
  ed0a0b69
23 Jun, 2025 7 commits
- ci: Revert openai_server related tests in AMD suites (#7449) · fa42e419
  Chang Su authored Jun 23, 2025
  
  fa42e419
- Support weight loading without mmap (#7469) · e5afb88b
  Yuhong Guo authored Jun 24, 2025
  
  e5afb88b
- Quick fix for DeepGemm requant to also cover MTP. (#7378) · e5ddeb04
  Charles Chen authored Jun 23, 2025
  
  e5ddeb04
- [perf] slightly imporve DeepSeek-R1-FP4 TP8 (#7481) · bdbb8d00
  JieXin Liang authored Jun 24, 2025
  
  bdbb8d00
- kvcache io kernels and test case (#7382) · 34c3f9b2
  Zhiqiang Xie authored Jun 23, 2025
  
  34c3f9b2
- update mooncake in dockerfile (#7480) · 76139bfb
  Liangsheng Yin authored Jun 24, 2025
  
  76139bfb
- Fix dtype for idle input in spec decoding (#7456) · f8d48fd3
  Cheng Wan authored Jun 23, 2025
  
  f8d48fd3