Commits · 09938e1f82334a98c124ae3352d893480963775f · change / sglang · GitLab

05 Nov, 2025 4 commits
- chore: bump SGLang version to 0.5.4.post3 (#12639) · 09938e1f
  sglang-bot authored Nov 05, 2025
  
  09938e1f
- Register allgather/reducescatter buffers with symm memory (#12572) · 23407983
  Nicolas Castet authored Nov 04, 2025
  
  23407983
- [fix] Handle escaped characters in GLM tool call parser to prevent double serialization (#12456) · 44da7377
  soaringk authored Nov 05, 2025
  
  44da7377
- Revert "Enable memory saver for hybrid model" (#12648) · d22d0447
  Baizhou Zhang authored Nov 04, 2025
  
  d22d0447
04 Nov, 2025 24 commits
- [NVIDIA] Fix wrong symmetric sizes for fp4 cases (#12640) · 34f7564d
  Kaixi Hou authored Nov 04, 2025
  
  34f7564d
- [Bug] Fix NSA Backend KV-Buffer Shape Mismatch in DeepSeek-V3.2 (#12645) · 1cfbbc42
  Johnsonms authored Nov 04, 2025
  
  1cfbbc42
- [Auto Sync] Update scheduler_metrics_mixin.py, collector.py (20251104) (#12647) · 55dfb539
  Lianmin Zheng authored Nov 04, 2025
```
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: SangBin Cho <rkooo567@gmail.com>
```
  55dfb539
- [hotfix] Fix deepep w4a8 bug (#12642) · 42889acb
  Baizhou Zhang authored Nov 04, 2025
  
  42889acb
- fix: Lazy import mooncake-ep to fix extra gpu contexts being created (#12641) · 211f4070
  Trevor Morris authored Nov 04, 2025
  
  211f4070
- Fix `output_ids` inconsistency (#12628) · befa41a1
  Liangsheng Yin authored Nov 05, 2025
  
  befa41a1
- Add io struct naming check back (#12634) · 30b26ee9
  Liangsheng Yin authored Nov 05, 2025
  
  30b26ee9
- [Test] Merge all constrained decoding tests. (#12633) · aa797d01
  Liangsheng Yin authored Nov 05, 2025
  
  aa797d01
- Fix skip layer in get_quant_method (#12632) · 7cee07a0
  Ke Bao authored Nov 04, 2025
  
  7cee07a0
- [HotFix] Disable torch dynamo for mrope_triton kernel (#12593) · bb517fe3
  Yuan Luo authored Nov 04, 2025
```
Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>
```
  bb517fe3
- Ensure GPU work is finished when release memory occupation call is finished (#12592) · ff0b64e1
  fzyzcjy authored Nov 04, 2025
  
  ff0b64e1
- [sepc-v2] Fix imcompatibility with constrained decoding (#12615) · 0678beaa
  Liangsheng Yin authored Nov 04, 2025
  
  0678beaa
- [Deterministic] Optimize bmm_batch_invariant op (#12522) · c14cc47e
  Minglei Zhu authored Nov 04, 2025
  
  c14cc47e
- Add --speculative-moe-runner-backend server arg (#10183) · dbcf85b7
  Trevor Morris authored Nov 04, 2025
  
  dbcf85b7
- feat: limit peak memory usage when computing logprobs (#6318) · d5fa019c
  Zhao Chen authored Nov 04, 2025
```
Signed-off-by: Zhao Chen <zhaochen.zju@gmail.com>
Co-authored-by: 赵晨阳 <zhaochen20@outlook.com>
```
  d5fa019c
- Enable memory saver for hybrid model (#11974) · 173e0f70
  Junrong Lin authored Nov 04, 2025
  
  173e0f70
- Improve the metrics for PD (#12580) · f600866a
  Lianmin Zheng authored Nov 03, 2025
```
Co-authored-by: Kan Wu <wukanustc@gmail.com>
Co-authored-by: cctry <shiyang@x.ai>
```
  f600866a
- fix: respect `--ignore-eos` in PD case for benchmarking (#12597) · 93be7e86
  ishandhanani authored Nov 03, 2025
  
  93be7e86
- Tiny fix ExpertDistributionReq error (#11760) · 60b0754c
  fzyzcjy authored Nov 04, 2025
  
  60b0754c
- test: support return logprobs in bench_offline_throughput test (#12462) · 0b24af4d
  Zhao Chen authored Nov 04, 2025
```
Signed-off-by: Zhao Chen <zhaochen.zju@gmail.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
```
  0b24af4d
- [Qwen3 VL] Add LoRA support for Qwen 3 VL (#12165) · a209fb05
  Jonah Bernard authored Nov 03, 2025
  
  a209fb05
- [GDN/SWA] mamba and swa radix cache edge case fix (#12111) · 48d6bea1
  Hanming Lu authored Nov 03, 2025
```
Co-authored-by: yizhang2077 <1109276519@qq.com>
```
  48d6bea1
- Enable mixed type LayerNorm kernel for NSA indexer (#12044) · e607850f
  akhilg-nv authored Nov 03, 2025
  
  e607850f
- Remove the dependency of nccl.h in symmetric memory (#12571) · 243c064d
  Lianmin Zheng authored Nov 03, 2025
  
  243c064d
03 Nov, 2025 12 commits
- update usage of `trtllm_fp8_per_tensor_scale_moe` (#12569) · d31d48b3
  b8zhong authored Nov 03, 2025
  
  d31d48b3
- Super tiny dump server info such as args in bench for post analysis (#12550) · 88342607
  fzyzcjy authored Nov 04, 2025
  
  88342607
- Super tiny allow profile activities in bench_serving (#12549) · fd7a72d6
  fzyzcjy authored Nov 04, 2025
  
  fd7a72d6
- tiny optimize for bench serving (#12553) · 21a8fa16
  Yi Zhang authored Nov 04, 2025
  
  21a8fa16
- Reduce the overhead of nccl symmetric memory (#12524) · 7a21d8b2
  Lianmin Zheng authored Nov 03, 2025
```
Co-authored-by: Nicolas Castet <ncastet@nvidia.com>
```
  7a21d8b2
- [Test] Add parameters to SRTRunner (#12227) · 6ef23b98
  Jonah Bernard authored Nov 03, 2025
  
  6ef23b98
- Fix error when calling quantization (#12548) · 385599cb
  fzyzcjy authored Nov 04, 2025
  
  385599cb
- fix: fix the bug which leads qwen2_5_vl to crash with mixed_chunk (#11330) · 952fbe47
  Yueyang Pan authored Nov 03, 2025
```
Signed-off-by: PanJason <pyyjason@gmail.com>
Co-authored-by: Xinyuan Tong <115166877+JustinTong0323@users.noreply.github.com>
Co-authored-by: Yuan Luo <yuan.luo@hotmail.com>
```
  952fbe47
- [hot-fix] Fix broken CI (#12564) · edb25693
  Liangsheng Yin authored Nov 04, 2025
  
  edb25693
- [spec v2] Fix output repetition by speculative sampling error (#12561) · 3529c061
  Liangsheng Yin authored Nov 03, 2025
  
  3529c061
- Conditionally recapture cuda graph after model weight update from disk (#12060) · ffb32a85
  harrisonlimh authored Nov 03, 2025
  
  ffb32a85
- fix: Fix KTransformers hybrid inference with int8 quantization and format (#12536) · 14d80648
  Atream authored Nov 03, 2025
  
  14d80648