Commit Graph

Select branches

Hide Pull Requests

dsv41-pd

dsv41-pd-visioncp

main

Mono Color

• a4db563c87 [hisparse]: update user guide (#26249) Zhanghengand晟海 2026-05-25 17:54:55 +08:00
• bc8d64bf36 [CI] Align score threshold in dsv4 disaggregation test (#26268) Shangming Cai 2026-05-25 17:48:45 +08:00
• a77449f86d [perf][spec decoding] Skip full-vocab softmax in EAGLE draft when topk == 1 (#26235) Qiaolin Yu 2026-05-25 02:06:48 -07:00
• 7c04b9e942 fix(docker): generate Cargo.lock in chef stage for sgl-router build (#26279) Kangyan-ZhouandClaude Opus 4.7 2026-05-25 16:54:47 +08:00
• 7f2829af39 chore: bump mooncake version to 0.3.11.post1 (#25989) Shangming Cai 2026-05-25 16:51:03 +08:00
• 0942011665 [NPU] Add torchaudio dependency for NPU platform (#26267) xdtbynd 2026-05-25 16:30:12 +08:00
• 81704ad602 ci: add nightly Docker workflow for experimental sgl-router (#26273) Kangyan-ZhouandClaude Opus 4.7 2026-05-25 16:19:07 +08:00
• e27d4fb70f [Perf][Qwen3.5] Add case 512 to topkGatingSoftmaxKernelLauncher, (#25775) Jincong Chen 2026-05-25 16:08:21 +08:00
• b0cf01eb85 Lazy-load speculative-naming via skill instead of always-on rule (#26270) fzyzcjy 2026-05-25 15:58:29 +08:00
• 6e8fe176be sgl-router: experimental Rust HTTP router for SGLang worker pools (#25851) Kangyan-ZhouandClaude Opus 4.7 2026-05-25 15:34:05 +08:00
• aae04b1241 📝 docs(diffusion): add MXFP4 quantization docs (#25904) Junlin Wu 2026-05-25 15:24:30 +08:00
• ca029e816b Fix missing idle-batch handling in prepare_mlp_sync_batch_raw (#25404) Yuhui Liang 2026-05-25 14:53:54 +08:00
• 87e69d57c4 [lora] Fix overlap loading for cancelled requests (#25413) Erik Wijmans 2026-05-24 23:18:36 -07:00
• 2bd3ac0b5d [XPU] fix correctness issue of GDN triton kernel for XPU (#26065) Xia WeiwenandMa Mingfei 2026-05-24 22:18:11 -07:00
• ec6fcb93cb [perf][spec decoding] Skip common_template in TRTLLMMLAMultiStepDraftBackend init (#26241) Qiaolin Yu 2026-05-24 21:36:16 -07:00
• e86fdf3a3c [Bug Fix][HiCache] TreeNode.get_prefix_hash_values @lru_cache can return mutated list (#26177) Ethan ZHUandZhangheng 2026-05-25 11:15:10 +08:00
• 821d5f4a5b [CPU] add faster KV-cache writes (#25874) Ma Mingfei 2026-05-25 10:28:52 +08:00
• e1463bb2c2 [VLM] try to reuse precomputed padded input ids in scheduler instead of padding (#26097) Mick 2026-05-25 10:27:04 +08:00
• de3f6fb02e Fix attr err (#25856) hanwlax 2026-05-25 10:26:19 +08:00
• 850887dc63 [Spec] fix EAGLE v2 verify metadata init order on non-cuda-graph path (#26244) Liangsheng Yin 2026-05-24 18:49:33 -07:00
• 64e2b54a8f [VLM] feat: accept grid_thws from preprocessed metadata for kimi (#26149) Mick 2026-05-25 09:10:25 +08:00
• 72c1582d4e [VLM] fix: fix only the grids from last split mm item is collected for qwen-vl (#26094) Mick 2026-05-25 09:09:46 +08:00
• ed179bf9b2 [dsv4] fix multi-step draft on non-cuda-graph path (#26239) Liangsheng Yin 2026-05-24 17:04:18 -07:00
• d7e3e54148 [Test] split test/registered/distributed/ into topic folders (#26240) Liangsheng Yin 2026-05-24 17:02:07 -07:00
• 85471d253d Add --disable-attn-tp-gather opt-out for model-managed SP (#26047) Ming Yang 2026-05-24 15:53:51 -07:00
• 93fa577bb9 Clean up server startup log noise (#26205) Lianmin Zheng 2026-05-24 14:35:15 -07:00
• 030bd5d3ed [Test] test_session_latency: assert streaming tail/head stability (#26230) Liangsheng Yin 2026-05-24 14:06:41 -07:00
• fd94bd30b8 [Intel GPU] DeepSeek V4 2/N: Fix tvm ffi import (#26118) Polisetty V R K Jyothendra Varma 2026-05-25 01:29:49 +05:30
• 44922de48a fix(swa): downgrade translate_loc_from_full_to_swa key-change log from warning to debug (#26225) Cheng WanandCheng Wan 2026-05-24 11:37:46 -07:00
• 7f45bcdd2a [dsv4] support eplb (#25948) Siyuan Chenandxutizhou 2026-05-25 01:09:41 +08:00
• 5c3775823e [diffusion] chore: use model-aware vae channels_last_3d policy (#26214) Mick 2026-05-25 00:25:34 +08:00
• 36eb72bf12 [UnifiedTree] fix: backup SWA-split parent before child under write-through (#25065) shuwenn 2026-05-25 00:21:57 +08:00
• 9d50cd9742 [observability] add ServerArgs.stat_loggers for pluggable metrics backend (#24610) Dongjun Na 2026-05-24 23:41:28 +09:00
• b6f71d5850 [VLM] avoid extra cuda-ipc staging for preprocessed input (#26096) Mick 2026-05-24 19:48:06 +08:00
• 6447596501 [VLM] feat: replace small H2D calls with a single one for qwen-vl (#26167) Mick 2026-05-24 18:27:17 +08:00
• 0b65588c18 [diffusion] Clean up VSA attention hot path (#25514) Xiaoyu Zhang 2026-05-24 16:46:03 +08:00
• 4c2b32bfbf [VLM] accept precomputed multimodal metadata (#26101) Mick 2026-05-24 15:43:21 +08:00
• 826a4de062 [srt] store req input ids as arrays (#26165) Mick 2026-05-24 15:09:36 +08:00
• d6d9f12444 [VLM] adopt simplified get_rope_index for image-only requests (#26100) Mick 2026-05-24 11:51:24 +08:00
• af8f66940e [AMD] Dsv4/pr1 fix run time issue (#25898) +212 people 2026-05-24 07:04:14 +08:00
• 982f67d9a6 Suppress cutlass-dsl noisy warning (#26169) Qiaolin Yu 2026-05-23 13:19:14 -07:00
• 2de74035a5 [FIX][2/2] fix step3-vl/deepseek-ocr image processor error (#25403) 3 people 2026-05-24 01:36:39 +08:00
• a5a64a311a [Spec] trtllm mha supports overlap plan stream (#25925) Hanming Lu 2026-05-23 03:25:57 -07:00
• cb7b57955d fix tokenspeed_mla attn kernel jit (#26170) Qiaolin Yu 2026-05-23 03:24:33 -07:00
• 89ff2bc111 [bug fix] Fix 3 issues when using Gemma4 MTP (#26026) Charles Chen 2026-05-23 03:16:47 -07:00
• b0ce16d0c5 [CP] 1/N: Support MLA Prefill Context Parallel (#23292) Khoa PhamandClaude Opus 4.7 2026-05-23 03:07:09 -07:00
• 81cd338fcc [docs] DeepSeek-V4 cookbook: balanced MegaMoE cap, H200 Pro FP4 mem-frac, nsa-* compat, PD-disagg fixes (#26164) zijiexiaandClaude Opus 4.7 2026-05-23 02:42:41 -07:00
• 1e59ed7443 compile _resolve_spec_extras gather kernels (#26129) Liangsheng Yin 2026-05-23 02:34:41 -07:00
• 83a18e687d Revert "[refactor] unify cuda-graph capture/replay across attention backends (#26134)" (#26166) Cheng WanandCheng Wan 2026-05-23 02:32:08 -07:00
• 8c78424701 Reduce excessively long logs caused by transformer version updates. (#26033) liuxianglong17 2026-05-23 17:12:08 +08:00
• 774b29dade [VLM] feat: early-return in mm processor if the input is preprocessed (#26117) Mick 2026-05-23 17:02:44 +08:00
• 19b60a4f9e [VLM] reuse pretokenized ids from preprocessed input for qwen-vl (#26116) Mick 2026-05-23 16:01:04 +08:00
• 8b9fb13c4a [BugFix][EPD] adapt for qwen3.5-mtp & del duplicated logs (#24144) Zheng Wengang 2026-05-23 15:32:57 +08:00
• 5964d30233 fix(swa): eliminate spurious translate_loc_from_full_to_swa warning in BCG and CG paths (#26152) Cheng WanandCheng Wan 2026-05-23 00:01:12 -07:00
• 75427c9ca4 Route concat MLA to JIT and remove unused downcast (#25843) Xiaoyu Zhang 2026-05-23 14:30:43 +08:00
• fd3e11973b [AMD][aiter] Fix cuda_graph_kv_indices OOB under page_size>1 (#24587) xiaobochen-amdandfanxingran 2026-05-23 14:19:46 +08:00
• c69844f043 [NPU]Ascend NPU Performance Profiling Guide and Ascend NPU Operator Development Guide (#26069) longxin9715 2026-05-23 10:50:56 +08:00
• 7b7f1067bd Add non-MTP DSV4 test coverage (#26141) Baizhou Zhangandyhyang201 2026-05-22 19:43:17 -07:00
• a241659d18 [PD] Consolidate shared logic into common backend (#25979) Shangming Cai 2026-05-23 10:41:20 +08:00
• c8cea6d4aa [diffusion] feat: auto-select vae channels_last_3d (#26121) Mick 2026-05-23 10:20:30 +08:00
• 629b6c6a85 correct allreduce fusion and dummy_run alignment in SCATTERED MLP mode (moe_dense_tp_size=1) (#19918) weireweire 2026-05-23 10:18:25 +08:00
• d226f75669 [refactor] unify cuda-graph capture/replay across attention backends (#26134) 3 people 2026-05-22 18:51:13 -07:00
• 208397affc [RL] [Spec v2] Use stop-aware seqlen for returned topk metadata (#26126) Yueming Yuan 2026-05-22 18:13:16 -07:00
• c112f7623a Skip init_mha_chunk_metadata in trtllm_mla when not needed (#26017) Qiaolin Yu 2026-05-22 16:34:16 -07:00
• cadfa2d025 Support piecewise CUDA graph with NSA (#23351) nvjullin 2026-05-23 05:39:50 +08:00
• 2df9e8b4b3 [perf] DeepSeekV3: drop redundant FP32 upcasts in trtllm MoE paths (#25189) maocheng23 2026-05-22 14:23:57 -07:00
• b73278e4e0 [Fix]: BCG support for RadixLinearAttention (Qwen3.5 / linear-attn hybrid models) (#25110) Minglei Zhu 2026-05-22 13:30:35 -07:00
• 763174fa6c FutureMap: debug-assert that gather sees a stashed value (#26108) Liangsheng Yin 2026-05-22 13:10:57 -07:00
• 085777210c feat(kv-events): expose structured KV-event publisher block on /server_info (#25844) Kangyan-ZhouandClaude Opus 4.7 2026-05-23 01:59:45 +08:00
• 06c23d55b5 perf: migrate Req token-id storage to array.array('q') in Scheduler (#25098) Jialin Ouyangandjialino 2026-05-22 10:51:07 -07:00
• 5e9bd21979 fix(serving_chat): catch TypeError from tojson on Jinja2 Undefined variables (#20700) fanghao 2026-05-23 00:43:26 +08:00
• b801a27ce6 [diffusion] CI: disable torch compile in nightly comparison (#26120) Mick 2026-05-22 23:36:47 +08:00
• 9bbd519f34 [diffusion] fix: fix Wan channels_last_3d VAE decode corruption (#25985) HuangJi 2026-05-22 23:34:41 +08:00
• 6baa859a86 Add MiniCPM5 tool call parser for XML-style function calls (#25600) zhangtao2-1andzhangtao 2026-05-22 23:09:40 +08:00
• f5ed2687ec [diffusion] CI: guard gt publishing from corrupt output (#26044) Mick 2026-05-22 19:07:20 +08:00
• 8c916a715c [diffusion] feat: support role-based component loading and stage affinity (#25168) Heyang Huang 2026-05-22 18:50:23 +08:00
• e1dcbca220 [FIX][1/2] fix step3-vl/deepseek-ocr image processor error (#24701) kousakawangandwanghanpei 2026-05-22 18:39:47 +08:00
• 80680dc3fe [Intel GPU] 1/N Fix tilelang import in deepseek v4 rope as optional (#25128) Polisetty V R K Jyothendra Varma 2026-05-22 15:53:18 +05:30
• d4082eab4d [CI] pr-test-extra: add run_all_tests to workflow_dispatch inputs (#26110) Kangyan-ZhouandClaude Opus 4.7 2026-05-22 18:03:39 +08:00
• 10751a4f0c drop FutureIndices wrapper class (#26085) Liangsheng Yin 2026-05-22 02:44:42 -07:00
• bd6c7e713c [fix] Fallback DeepGEMM activation for unsupported shapes (#26025) Yueming Yuan 2026-05-22 02:01:40 -07:00
• 4486a339a3 [CI] bot-cherry-pick: remove concurrency group to enable batch dispatch (#26074) Kangyan-ZhouandClaude Opus 4.7 2026-05-22 16:27:46 +08:00
• cf5f496183 [MUSA][22/N] ci(musa): repack wheels with +musa metadata, refine path filters, sync multimodal tests, and add nightly workflow (#25074) johnnycxmandxinyue.fan 2026-05-22 16:24:25 +08:00
• 16d049f898 Add sglang-cherrypick skill for batching bot-cherry-pick dispatches (#26068) Kangyan-ZhouandClaude Opus 4.7 2026-05-22 16:20:06 +08:00
• 63ecf2f62b [CI] Drop unused 'environment: prod' from bot-cherry-pick job (#26067) Kangyan-Zhou 2026-05-22 16:16:07 +08:00
• 06836aa38a update code owner (#26060) siyu 2026-05-22 16:13:28 +08:00
• 6339295556 [XPU] add apache-tvm-ffi dependency (#26053) jiayisunx 2026-05-22 16:09:08 +08:00
• 88a37d7405 [docs] DeepSeek-V4 cookbook: split Quantization axis, add H100 SGLang FP8 (#26057) zijiexiaandClaude Opus 4.7 2026-05-22 00:52:10 -07:00
• c9153da5dc Fix SWA double-free in disagg decode with MTP speculation (#25805) Yuhao YangandKe Bao 2026-05-22 15:28:54 +08:00
• a7555fc997 [diffusion] fix: fix MOVA DAC bf16 on ROCm (#25674) Chi McIsaac 2026-05-22 03:18:49 -04:00
• c4b6b5ea1e [core] step 2: drop seq_lens sentinel; SB maintains GPU as seq_lens_cpu mirror (#26020) Liangsheng Yin 2026-05-22 00:12:51 -07:00
• fa6f4dfb35 improve: combine vit calls for images from different reqs from one batch (#25910) Yuhao YangandYaochen Han 2026-05-22 14:57:41 +08:00
• 3f0814974c Fix disaggregation bootstrap server lifetime (#25982) fzyzcjy 2026-05-22 14:10:57 +08:00
• f829cafa3e [perf] skip add_special_tokens=False kwarg on chat-template tokenize for slow tokenizers (#25953) Khoa PhamandClaude Opus 4.7 2026-05-21 22:45:54 -07:00
• acb8310183 ci: self-heal $GITHUB_PATH/$GITHUB_ENV writes (#26037) Liangsheng Yin 2026-05-21 21:28:43 -07:00
• 610f55040d update npu codeowners (#26042) ybyang 2026-05-22 11:46:59 +08:00
• ae7c4226eb [diffusion] model: support FLUX.2-klein-base (#25661) Alex O. P. 2026-05-22 05:24:46 +02:00
• 8b473aa0bc [core] step 1: route non-spec seq_lens via FutureMap with per-mode bootstrap fixes (#25944) Liangsheng Yin 2026-05-21 20:15:51 -07:00
• cc77c36029 Bugfix: Qwen3-VL-MoE adapt encoder_only (#23220) jianzhao-xu 2026-05-22 11:12:13 +08:00
• 76b1efaffd Fix SMG service discovery Clippy lint (#26034) Mohammad Miadh Angkad 2026-05-22 10:51:13 +08:00