minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity
14,507 Commits 2 Branches 0 Tags
212c30d00899bf3f32c5de37ffa924318a67af25
Commit Graph
9803 Commits
Author SHA1 Message Date
Mohammad Miadh Angkad 212c30d008 [MoE Refactor] Centralize FlashInfer CUTLASS MoE runner (#28211) 2026-06-25 13:40:33 -07:00
cctry dbe9e3b706 [PD] Early-send cached-prefix KV overlapping uncached prefill forward (#29316) 2026-06-25 13:31:53 -07:00
Lianmin Zheng 118d6b2e5e [Cleanup] Style and type annotation improvements extracted from #28688 (#29224) 2026-06-25 12:51:15 -07:00
weireweire f9a3720e2b Skip empty non-idle output batches (#28504) 2026-06-26 02:40:01 +08:00
Xiaoyu Zhangjingyu-mlClaude Opus 4.8
 		52c32035eb [diffusion] Add Qwen-Image ModelOpt NVFP4 support (#28928) 
Co-authored-by: jingyu-ml <jingyux@nvidia.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-06-25 22:56:09 +08:00
Mick 890b38c211 [diffusion] doc: fix diffusion docs and cookbook drift (#29302) 2026-06-25 21:13:00 +08:00
Xiaoyu Zhang ddda4f9028 [Bugfix] Fix Ministral3 init argument forwarding (#29111) 2026-06-25 18:26:45 +08:00
Rita Brugarolas 3d3a7ec031 [AMD] fix(moe): correct fused shared-expert scaling on aiter/DeepEP path (mori all-to-all) (#28237) 
Signed-off-by: Rita Brugarolas Brufau <rita.brugarolasbrufau@amd.com>
 2026-06-25 01:56:30 -07:00
Rita Brugarolas de2d01c8da [AMD] Fuse shared-expert append + DeepEP remap into one Triton kernel (#28450) 
Signed-off-by: Rita Brugarolas Brufau <rita.brugarolasbrufau@amd.com>
 2026-06-25 01:22:57 -07:00
Yuhao Yang 4a8200565e Fused QK GemmaRMSNorm + RoPE + gate kernel for Qwen3.5 (#28320) 2026-06-25 15:58:54 +08:00
Liangsheng Yin 2812a3c93a [Spec] Unify spec/non-spec decode result handling and overlap relay-payload gating (#29225) 2026-06-25 00:44:38 -07:00
jiayisunx d717e926c1 fix(runner): prevent eager token buffer under-allocation (#28894) 2026-06-25 14:37:19 +08:00
Raiden MakotoandRaiden-Makoto 9215da2515 [AMD] [GLM5] skip redundant -inf pre-fill of HIP indexer MQA-logits (#28757) 
Co-authored-by: Raiden-Makoto <Raiden-Makoto@users.noreply.github.com>
 2026-06-24 23:21:39 -07:00
AndyLi429 72cac88022 [NPU] perf: precompute mamba conv-state track indices once per batch (#29105) 2026-06-25 14:04:56 +08:00
Yanbin JiangandYusheng Su 9fd6d0ec99 [LoRA] BF16 support + EP cuda-graph crash fix for experimental_sgl_trtllm MoE-LoRA (#28953) 
Co-authored-by: Yusheng Su <yushengsu.thu@gmail.com>
 2026-06-24 21:16:50 -07:00
wenxuewuhdandronnie_zheng a9270250c3 Dllm radix cache npu (#27144) 
Co-authored-by: ronnie_zheng <zl19940307@163.com>
 2026-06-25 06:44:03 +03:00
CrazyCoderandClaude Opus 4.8 c268a7edf5 [Bugfix] Fix inverted PP-disable condition for diffusion LLM inference (#27700) 
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-06-24 20:18:00 -07:00
Mohammad Miadh Angkad 8b7a1e908a Revert Gemma4 modelopt fp4 MoE backend change (#28347) 2026-06-24 19:49:09 -07:00
Mohammad Miadh Angkad 40439acd0a Sync backend docs with #29063 (#29233) 2026-06-24 19:48:34 -07:00
Liangsheng Yin 8314247d9d [Spec] Merge dflash triton kernels into a single dflash.py (#29228) 2026-06-24 18:57:15 -07:00
Fenglin Yuklhhhhhzhaochenyang20shuwennMikukuOvO
 		ae6f787637 [diffusion] rl: add sleep/wake support for diffusion engine (#22659) 
Co-authored-by: klhhhhh <1412841649@qq.com>
Co-authored-by: zhaochenyang20 <zhaochen20@outlook.com>
Co-authored-by: shuwenn <47200617+alphabetc1@users.noreply.github.com>
Co-authored-by: MikukuOvO <MikukuOvO@users.noreply.github.com>
 2026-06-25 09:37:27 +08:00
看海的人 67b2a9ed0c [NPU] Fix the DeepSeek-V2-Coder model accuracy issue (#29042) 2026-06-25 09:14:32 +08:00
chx96642264 4e155ecc31 [NPU] adaptation to support operator FA3 in deterministic inference on NPU. (#26724) 2026-06-25 09:12:57 +08:00
Liangsheng Yin c7734e6871 [Spec] Dissolve EagleDraftInputV2Mixin so spec-info dataclasses hold data only (#29220) 2026-06-24 18:08:30 -07:00
fzyzcjy 2b64fc7a2c Remove Req.extend_logprob_start_len field and make it pure (#27625) 2026-06-25 09:06:14 +08:00
fzyzcjy 7002a37ea1 Inline extend_range accessors and remove the extend_input_len/fill_len properties (#27611) 2026-06-25 08:56:44 +08:00
cen121212 9305d10099 [NPU] adapt_fused_rope_qk_mqa_optimize (#28872) 2026-06-25 08:55:52 +08:00
fzyzcjy d0524d6433 Avoid scattered assignment of extend_input_len and fill_len by merging them into Req.extend_range (#27610) 2026-06-25 08:33:28 +08:00
Lianmin Zheng 6c839368e0 [Cleanup] IPC struct renames, better typing, and SenderWrapper removal (#29214) 2026-06-24 17:25:24 -07:00
fzyzcjy 563c3418a7 Avoid dual semantics of extend_input_len by computing the candidate on the fly (#27616) 2026-06-25 08:17:37 +08:00
Brayden ZhongandBrayden Zhong 5d4e63d49e Sync the changes in #23402 (#29063) 
Co-authored-by: Brayden Zhong <brayden@radixark.ai>
 2026-06-24 17:01:02 -07:00
Brayden ZhongandBrayden Zhong f82addd4a8 Support online MXFP8 quantization for ungated MoE (#27939) 
Co-authored-by: Brayden Zhong <brayden@radixark.ai>
 2026-06-24 16:58:48 -07:00
Jiajun LiandClaude Opus 4.8 e4bf0043fe [fix] Skip routed expert capture for draft model under spec v2 (#26980) 
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-06-24 16:56:07 -07:00
Lianmin ZhengandYinghai Lu 7e63feee6f Add scheduler metrics extension hooks (#29207) 
Co-authored-by: Yinghai Lu <yinghai@meta.com>
 2026-06-24 15:50:05 -07:00
Qiaolin Yu e26bceb81e [perf] simplify _apply_cuda_graph_metadata for draft extend in trtllm_mla backend (#29077) 2026-06-24 15:45:26 -07:00
Liangsheng Yin 85a3f1f75c [Spec] Unify the overlap stash relay behind a RelayPayload dataclass (#29124) 2026-06-24 15:36:54 -07:00
Liangsheng Yin c6822f81b1 [Spec] Make the overlap bonus-token relay unconditional (#29122) 2026-06-24 15:34:02 -07:00
Liangsheng Yin d3dd184e3e [Spec] Fold DFlash verified_id into the shared bonus_tokens relay channel (#29118) 2026-06-24 15:30:29 -07:00
Brayden ZhongandBrayden Zhong 3c95a87b66 Fix the CuDNN failure on bmm_fp8 when two libcudart.so exists. (#29201) 
Co-authored-by: Brayden Zhong <brayden@radixark.ai>
 2026-06-24 15:26:46 -07:00
Lianmin ZhengandRain Jiang fd87a85388 Convert SamplingParams to msgspec Struct (#29198) 
Co-authored-by: Rain Jiang <96632942+rainj-me@users.noreply.github.com>
 2026-06-24 13:42:45 -07:00
Kaixi d5e9176f65 [BCG][GLM5] perf: BCG support and prefill enhancements (#27053) 2026-06-24 13:13:35 -07:00
Brayden ZhongandBrayden Zhong d46afbf8b4 Fix nightly CI test for GLM-4.6 + B200 (#28749) 
Co-authored-by: Brayden Zhong <brayden@radixark.ai>
 2026-06-24 12:34:11 -07:00
Qiaolin Yu c01ad10d16 [perf] tiny optimize select_index op for draft extend (#29078) 2026-06-24 12:27:13 -07:00
Lianmin Zheng d5c566e59b Extract profile request cleanups (#29098) 2026-06-24 11:22:58 -07:00
Lianmin ZhengandCheng Wan d6aacd2801 Handle input-embed-only batches in eager runner (#29121) 
Co-authored-by: Cheng Wan <54331508+ch-wan@users.noreply.github.com>
 2026-06-24 10:33:52 -07:00
cctryandcctry 76db6c9d9e [Refactor] Share CUDA graph memory pool across prefill and decode (#28973) 
Co-authored-by: cctry <cctry@fb.com>
 2026-06-24 10:18:20 -07:00
Mick 4a4f063b79 [diffusion] fix: paint multiview vae must follow unit dtype (#29041) 2026-06-24 23:54:01 +08:00
qiaozpandishandhanani 03773ae35b [HiCache] Add NIXL FILE cache cleaner (#28258) 
Co-authored-by: ishandhanani <82981111+ishandhanani@users.noreply.github.com>
 2026-06-24 08:13:55 -07:00
Mick 09b808ab7e [diffusion] fix: fix Qwen-Image-Layered latent shape (#28832) 2026-06-24 20:57:54 +08:00
nvjullin 84a7a84018 [PD] Fix data race in NixlKVManager for NIXL backend (#28897) 2026-06-24 18:11:01 +08:00