minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity
13,074 Commits 2 Branches 0 Tags
b6f71d5850eadd97de8a7d569e7271a58a9ffa2e
Commit Graph
13074 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
Mick b6f71d5850 [VLM] avoid extra cuda-ipc staging for preprocessed input (#26096) 2026-05-24 19:48:06 +08:00
Mick 6447596501 [VLM] feat: replace small H2D calls with a single one for qwen-vl (#26167) 2026-05-24 18:27:17 +08:00
Xiaoyu Zhang 0b65588c18 [diffusion] Clean up VSA attention hot path (#25514) 2026-05-24 16:46:03 +08:00
Mick 4c2b32bfbf [VLM] accept precomputed multimodal metadata (#26101) 2026-05-24 15:43:21 +08:00
Mick 826a4de062 [srt] store req input ids as arrays (#26165) 2026-05-24 15:09:36 +08:00
Mick d6d9f12444 [VLM] adopt simplified get_rope_index for image-only requests (#26100) 2026-05-24 11:51:24 +08:00
+2
kkwunhuangThomas WangXinyi SongHaiShawamd-danli103Lin, SogaRaiden-MakotoHubert Luyichiche@amd.comyctseng0211Bingxu Chen
 		af8f66940e [AMD] Dsv4/pr1 fix run time issue (#25898) 
Co-authored-by: wunhuang <wunhuang@amd.com>
Co-authored-by: Thomas Wang <1am9trash@gmail.com>
Co-authored-by: Xinyi Song <86638975+RolaoDenthu@users.noreply.github.com>
Co-authored-by: HaiShaw <hixiao@gmail.com>
Co-authored-by: amd-danli103 <danli103@amd.com>
Co-authored-by: Lin, Soga <soga.lin@amd.com>
Co-authored-by: Raiden-Makoto <Raiden-Makoto@users.noreply.github.com>
Co-authored-by: Hubert Lu <55214931+hubertlu-tw@users.noreply.github.com>
Co-authored-by: yichiche@amd.com <jacky.cheng>
Co-authored-by: yctseng0211 <yctseng@amd.com>
Co-authored-by: Bingxu Chen <bingxche@amd.com>
 2026-05-23 16:04:14 -07:00
Qiaolin Yu 982f67d9a6 Suppress cutlass-dsl noisy warning (#26169) 2026-05-23 13:19:14 -07:00
kousakawangkousakawanggemini-code-assist[bot]
 		2de74035a5 [FIX][2/2] fix step3-vl/deepseek-ocr image processor error (#25403) 
Co-authored-by: kousakawang <wanghanpei@bytedance.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
 2026-05-24 01:36:39 +08:00
Hanming Lu a5a64a311a [Spec] trtllm mha supports overlap plan stream (#25925) 2026-05-23 03:25:57 -07:00
Qiaolin Yu cb7b57955d fix tokenspeed_mla attn kernel jit (#26170) 2026-05-23 03:24:33 -07:00
Charles Chen 89ff2bc111 [bug fix] Fix 3 issues when using Gemma4 MTP (#26026) 2026-05-23 03:16:47 -07:00
Khoa PhamandClaude Opus 4.7 b0ce16d0c5 [CP] 1/N: Support MLA Prefill Context Parallel (#23292) 
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-23 03:07:09 -07:00
zijiexiaandClaude Opus 4.7 81cd338fcc [docs] DeepSeek-V4 cookbook: balanced MegaMoE cap, H200 Pro FP4 mem-frac, nsa-* compat, PD-disagg fixes (#26164) 
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-23 02:42:41 -07:00
Liangsheng Yin 1e59ed7443 compile _resolve_spec_extras gather kernels (#26129) 2026-05-23 02:34:41 -07:00
Cheng WanandCheng Wan 83a18e687d Revert "[refactor] unify cuda-graph capture/replay across attention backends (#26134)" (#26166) 
Co-authored-by: Cheng Wan <cheng.wan@radixark.ai>
 2026-05-23 02:32:08 -07:00
liuxianglong17 8c78424701 Reduce excessively long logs caused by transformer version updates. (#26033) 2026-05-23 17:12:08 +08:00
Mick 774b29dade [VLM] feat: early-return in mm processor if the input is preprocessed (#26117) 2026-05-23 17:02:44 +08:00
Mick 19b60a4f9e [VLM] reuse pretokenized ids from preprocessed input for qwen-vl (#26116) 2026-05-23 16:01:04 +08:00
Zheng Wengang 8b9fb13c4a [BugFix][EPD] adapt for qwen3.5-mtp & del duplicated logs (#24144) 2026-05-23 15:32:57 +08:00
Cheng WanandCheng Wan 5964d30233 fix(swa): eliminate spurious translate_loc_from_full_to_swa warning in BCG and CG paths (#26152) 
Co-authored-by: Cheng Wan <cheng.wan@radixark.ai>
 2026-05-23 00:01:12 -07:00
Xiaoyu Zhang 75427c9ca4 Route concat MLA to JIT and remove unused downcast (#25843) 2026-05-23 14:30:43 +08:00
xiaobochen-amdandfanxingran fd3e11973b [AMD][aiter] Fix cuda_graph_kv_indices OOB under page_size>1 (#24587) 
Co-authored-by: fanxingran <fanxingran@amd.com>
 2026-05-22 23:19:46 -07:00
longxin9715 c69844f043 [NPU]Ascend NPU Performance Profiling Guide and Ascend NPU Operator Development Guide (#26069) 2026-05-23 10:50:56 +08:00
Baizhou Zhangandyhyang201 7b7f1067bd Add non-MTP DSV4 test coverage (#26141) 
Co-authored-by: yhyang201 <yhyang201@gmail.com>
 2026-05-22 19:43:17 -07:00
Shangming Cai a241659d18 [PD] Consolidate shared logic into common backend (#25979) 
Signed-off-by: Shangming Cai <csmthu@gmail.com>
 2026-05-23 10:41:20 +08:00
Mick c8cea6d4aa [diffusion] feat: auto-select vae channels_last_3d (#26121) 2026-05-23 10:20:30 +08:00
weireweire 629b6c6a85 correct allreduce fusion and dummy_run alignment in SCATTERED MLP mode (moe_dense_tp_size=1) (#19918) 2026-05-22 19:18:25 -07:00
Cheng WanCheng WanClaude Sonnet 4.6
 		d226f75669 [refactor] unify cuda-graph capture/replay across attention backends (#26134) 
Co-authored-by: Cheng Wan <cheng.wan@radixark.ai>
Co-authored-by: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>
 2026-05-22 18:51:13 -07:00
Yueming Yuan 208397affc [RL] [Spec v2] Use stop-aware seqlen for returned topk metadata (#26126) 2026-05-22 18:13:16 -07:00
Qiaolin Yu c112f7623a Skip init_mha_chunk_metadata in trtllm_mla when not needed (#26017) 2026-05-22 16:34:16 -07:00
nvjullin cadfa2d025 Support piecewise CUDA graph with NSA (#23351) 2026-05-22 14:39:50 -07:00
maocheng23 2df9e8b4b3 [perf] DeepSeekV3: drop redundant FP32 upcasts in trtllm MoE paths (#25189) 2026-05-22 14:23:57 -07:00
Minglei Zhu b73278e4e0 [Fix]: BCG support for RadixLinearAttention (Qwen3.5 / linear-attn hybrid models) (#25110) 2026-05-22 13:30:35 -07:00
Liangsheng Yin 763174fa6c FutureMap: debug-assert that gather sees a stashed value (#26108) 2026-05-22 13:10:57 -07:00
Kangyan-ZhouandClaude Opus 4.7 085777210c feat(kv-events): expose structured KV-event publisher block on /server_info (#25844) 
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-23 01:59:45 +08:00
Jialin Ouyangandjialino 06c23d55b5 perf: migrate Req token-id storage to array.array('q') in Scheduler (#25098) 
Co-authored-by: jialino <jialino@fb.com>
 2026-05-22 10:51:07 -07:00
fanghao 5e9bd21979 fix(serving_chat): catch TypeError from tojson on Jinja2 Undefined variables (#20700) 2026-05-22 16:43:26 +00:00
Mick b801a27ce6 [diffusion] CI: disable torch compile in nightly comparison (#26120) 2026-05-22 23:36:47 +08:00
HuangJi 9bbd519f34 [diffusion] fix: fix Wan channels_last_3d VAE decode corruption (#25985) 2026-05-22 23:34:41 +08:00
zhangtao2-1andzhangtao 6baa859a86 Add MiniCPM5 tool call parser for XML-style function calls (#25600) 
Co-authored-by: zhangtao <zhangtao2@modelbest.cn>
 2026-05-22 23:09:40 +08:00
Mick f5ed2687ec [diffusion] CI: guard gt publishing from corrupt output (#26044) 2026-05-22 19:07:20 +08:00
Heyang Huang 8c916a715c [diffusion] feat: support role-based component loading and stage affinity (#25168) 2026-05-22 18:50:23 +08:00
kousakawangandwanghanpei e1dcbca220 [FIX][1/2] fix step3-vl/deepseek-ocr image processor error (#24701) 
Co-authored-by: wanghanpei <wanghanpei@bytedance.com>
 2026-05-22 18:39:47 +08:00
Polisetty V R K Jyothendra Varma 80680dc3fe [Intel GPU] 1/N Fix tilelang import in deepseek v4 rope as optional (#25128) 2026-05-22 18:23:18 +08:00
Kangyan-ZhouandClaude Opus 4.7 d4082eab4d [CI] pr-test-extra: add run_all_tests to workflow_dispatch inputs (#26110) 
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
 2026-05-22 18:03:39 +08:00
Liangsheng Yin 10751a4f0c drop FutureIndices wrapper class (#26085) 2026-05-22 02:44:42 -07:00
Yueming Yuan bd6c7e713c [fix] Fallback DeepGEMM activation for unsupported shapes (#26025) 2026-05-22 09:01:40 +00:00
Kangyan-ZhouandClaude Opus 4.7 4486a339a3 [CI] bot-cherry-pick: remove concurrency group to enable batch dispatch (#26074) 
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
 2026-05-22 16:27:46 +08:00
johnnycxmandxinyue.fan cf5f496183 [MUSA][22/N] ci(musa): repack wheels with +musa metadata, refine path filters, sync multimodal tests, and add nightly workflow (#25074) 
Co-authored-by: xinyue.fan <xinyue.fan@mthreads.com>
 2026-05-22 16:24:25 +08:00