Commit Graph

Select branches

Hide Pull Requests

bot/update-est-time-20260928

dsv41-pd

dsv41-pd-visioncp

main

Mono Color

• c83ef4fdb6 use envs in server_args (#22994) Liangsheng Yin 2026-04-16 15:01:33 -07:00
• c0172aef6e [Ray] Bind scheduler actors to GPU-local NUMA node (#22989) Xinyu Zhangandxyuzh 2026-04-16 14:52:15 -07:00
• d430034bde [Ray] Support multi-replica serving by making scheduler actor names unique (#22917) Xinyu Zhang 2026-04-16 14:51:01 -07:00
• a87806a65f [misc] refine outdated comments for chain-style multi-layer MTP (#22996) Qiaolin Yu 2026-04-16 14:49:43 -07:00
• 12266cf953 [misc] update .github/CODEOWNERS (#22993) Qiaolin Yu 2026-04-16 14:19:41 -07:00
• 41258f874d [PD]feat(bench): add --fake-prefill flag for decode-only stress testing (#22973) ybyang 2026-04-17 04:57:55 +08:00
• 29f56cb230 CI: fix lint (#22991) Mick 2026-04-17 02:09:04 +08:00
• 0882f5c132 [Doc] correct the HTTP endpoint for stopping profiling in benchmark_and_profiling.md (#22523) Yujun Dong 2026-04-17 00:54:28 +08:00
• 71377deda7 [Docs] fix profiling endpoint (#22982) Zaire 2026-04-17 00:51:39 +08:00
• 082eaed0a4 test: fix flaky required function calling assertion (#22890) Xinyuan Tong 2026-04-16 17:44:26 +01:00
• 9da998a882 [diffusion] feat: disaggregated diffusion (#21701) Yuhao Yang 2026-04-16 23:51:32 +08:00
• 14bcdfca21 [HiSparse]: Adding e2e ut for hisparse (#22979) Zhangheng 2026-04-16 23:20:07 +08:00
• 78147306b7 [NPU] [DOC] Update npu best practice docs to match latest code (#22975) amote-i 2026-04-16 20:45:22 +08:00
• bbd8f9ba09 migrate CPU-only unit tests from openai_server to unit/ (#22965) Liangsheng Yin 2026-04-16 03:53:33 -07:00
• 62309f09db fix(loads): preserve include filtering after watching mode switch (#22959) Liangsheng Yin 2026-04-16 03:04:53 -07:00
• 03fef357a6 fix(loads): switch get_loads_communicator to watching mode (#22919) ybyang 2026-04-16 17:12:22 +08:00
• fbd6dc3565 fix: normalize tool message content for GLM5.1 chat template (#22595) ybyang 2026-04-16 16:48:38 +08:00
• aaa682346e [diffusion] model: Properly validate device for Mistral 3 attention (#22690) Aleksi Vesanto 2026-04-16 10:29:23 +03:00
• 1412e287bf [AMD][MoRI] bump MoRI to v1.1.0 (#22870) jhchouuu 2026-04-16 15:11:55 +08:00
• 35da90cb76 [misc] Configure logging before ServerArgs.__post_init__ (#22926) Lianmin Zheng 2026-04-15 23:53:15 -07:00
• 65bc839a5f [Fix] eagle/eagle3 speculative decoding conflicts with xgrammar in NPU (#20989) yuefeng Wuandgemini-code-assist[bot] 2026-04-16 14:34:23 +08:00
• c43716a357 [sgl] provide an option to send control req to all dp ranks rank0 (#22758) Bi Xue 2026-04-15 23:24:26 -07:00
• 3600465e81 [Bug Fix] Remove follow_bootstrap_room fast path in PD disaggregation DP rank resolution (#22901) Byron Hsu 2026-04-15 22:53:29 -07:00
• 2480cc2a16 docs: fix incorrect default max-payload-size in gateway config reference (#22923) hhwxw 2026-04-16 01:25:27 -04:00
• e7ad7c587a [EPD][VLM] Support Kimi VL EPD (#22490) LHXuuu 2026-04-16 12:40:02 +08:00
• 58c6b871b2 Remove compatibility restriction between Pipeline Parallelism and Mixed Chunked Prefill (#22920) CYYYC0310andcyy 2026-04-16 11:25:31 +08:00
• 34fef07a15 Upgrade transformers to 5.5.3 and refactor hf_transformers_utils into subpackage (#21569) Xinyuan Tong 2026-04-16 04:03:44 +01:00
• 14e122cdee [BugFix][RadixTree]:Fix stale eviction assertion in HiMambaRadixCache host eviction path (#22592) JINZandZhangheng 2026-04-16 10:49:30 +08:00
• 5ef67cee16 [AMD] Fix aiter import failure in ROCm Docker images (#22363) Vladimir (Vova) Vagaytsev 2026-04-16 03:40:05 +02:00
• b8794baa6d [Step3p5] Optimize allreduce in MoE layers (#22773) Yuhao Yang 2026-04-16 09:33:12 +08:00
• a4cf2ea128 streaming session: spec v2 bonus accounting + comprehensive test matrix (#22651) Liangsheng Yin 2026-04-15 17:12:41 -07:00
• ccff59254c Update .codespellrc (#22912) Lianmin Zheng 2026-04-15 16:25:55 -07:00
• 761259448d ci: re-enable fp8 nightly benchmark configs (#22910) ishandhananiandClaude Opus 4.6 2026-04-15 17:57:49 -05:00
• e8c6e5466c [Ray] Auto-create placement group in RayEngine when none is detected (#22898) Xinyu Zhang 2026-04-15 15:17:52 -07:00
• 0b1b07db72 [misc] fix ray folder lint (#22905) Qiaolin Yu 2026-04-15 15:08:18 -07:00
• f9792166c3 trim_overshoot: cap swa_evicted_seqlen + unit test (#22900) Liangsheng Yin 2026-04-15 15:05:35 -07:00
• 2b0f349927 ci: clarify srt-slurm issue filing for incompatible flag combos (#22903) ishandhananiandClaude Opus 4.6 2026-04-15 17:02:26 -05:00
• 13a2cd748d [Ray] Add data parallel (DP) and DP attention support to RayEngine (#21887) Xinyu Zhangandxyuzh 2026-04-15 15:00:48 -07:00
• 4927975427 [Score API] Add return_pooled_hidden_states to Scoring API for SequenceClassification / RewardModel (#22427) Sundara Raman Ramachandran 2026-04-15 14:58:56 -07:00
• 4e480d5785 Harden FlashInfer FP4 imports in standard dispatcher (#21776) Lee Nau 2026-04-15 14:54:49 -07:00
• 9497001b0c ci: add issue filing and suspect PR identification to log analyzer (#22899) ishandhananiandClaude Opus 4.6 2026-04-15 16:27:14 -05:00
• efc267ca29 streaming session: trim spec v2 overshoot in cache_finished_req (#22897) Liangsheng Yin 2026-04-15 14:15:46 -07:00
• f61c332cba ci: log analyzer (#22859) ishandhanani 2026-04-15 16:10:00 -05:00
• 43925d179d [Speculative] Fix Eagle3/DFLASH aux hidden state capture during CUDA graph init (#22836) Lianmin Zheng 2026-04-15 14:04:54 -07:00
• 32d9fe5a32 [lora] Speedup triton backend sgemm calls with better grid (#22386) Kurt Shuster 2026-04-15 16:47:07 -04:00
• 113d654152 [Fix] Fix accuracy bug in Flashmla sparse MLA kernel (#22723) Baizhou Zhang 2026-04-15 13:40:04 -07:00
• 28e915b474 [Bugfix] Preserve auto-detected quant_config for GLM NextN draft model (#22823) Jimmy Shong 2026-04-15 13:25:36 -07:00
• 8686f42acb [VLM] Enable per-image ViT cache and avoid TP CUDA context creation for Kimi-K2.5 (#22858) Yuhao Yang 2026-04-16 01:14:24 +08:00
• 7d7fdc1309 [HiCache]Fix CP support for hybrid model (#22782) huangtingweiandhzh0425 2026-04-15 23:50:29 +08:00
• 9e84f53785 [PD] Add a fallback to bypass rust dep for mini_lb (#21982) ybyang 2026-04-15 22:34:36 +08:00
• 695ab705cb [diffusion] quant: update modelopt quantization docs and CI coverage (#22772) Xiaoyu Zhang 2026-04-15 21:30:28 +08:00
• 80718492dd [diffusion] CI: reset thresholds (#22854) Mick 2026-04-15 21:11:00 +08:00
• 0a5c9728a1 [HiSparse][BugFix]: Fix the memory leak issue during health checks. (#22882) Zhangheng 2026-04-15 19:49:54 +08:00
• ce31934ca8 Streaming session: fix retract tail leak via _free_tail (#22862) Liangsheng Yin 2026-04-15 01:44:27 -07:00
• 3511c2deb4 [HiCache] Fix memory host free logic when share_indices_with_anchor enabled (#22767) huangtingweiandhzh0425 2026-04-15 16:31:18 +08:00
• aa78564e1a Refactor streaming session abort handling (#22790) Liangsheng Yin 2026-04-15 00:13:05 -07:00
• 45a83ffbe3 [NPU] Offloading docs update (#22860) jianzhao-xuandJianzhao Xu 2026-04-15 15:04:41 +08:00
• b2af34be54 [AMD] Optimize _append_shared_to_topk_output by a single fused Triton kernel for Qwen3.5 (#22844) Hubert LuandHaiShaw 2026-04-14 23:50:32 -07:00
• ada52e5972 [Docs] Move ptxas sm_103a workaround into For CUDA 13 section (#22852) Po-Han Huang (NVIDIA)andClaude Opus 4.6 2026-04-15 13:30:21 +08:00
• e95c2e73bd [diffusion] CI: refactor diffusion ci and reduce redundancy (#22810) Mick 2026-04-15 10:12:29 +08:00
• 47ac830c07 [diffusion] rl: support standalone rollout api, denoising environment backpass and sp-aligned log-prob for T2I post-training (#22604) 3 people 2026-04-14 19:10:38 -07:00
• 39c6bf730c [AMD][CI] Add GLM-5-MXFP4 accuracy and perf nightly tests for MI35x (#21773) Michael 2026-04-14 18:55:36 -07:00
• adb310b976 Cleanup server_args.py and minor code tidying (#22820) Lianmin Zheng 2026-04-14 18:52:41 -07:00
• ea05ea5abe [AMD] Enable share expert fusion with router experts for Qwen3.5 BF16 & FP8 (#20736) 3 people 2026-04-15 09:52:36 +08:00
• 46c8a597ef [VLM] fix LFM2-VL offline inference and GPU JPEG decode (#22448) Piotr Mazurek 2026-04-15 03:13:25 +02:00
• 2c9e76d333 ci: skip approval for nightly gb200 runs, keep for manual triggers (#22768) ishandhanani 2026-04-14 18:34:57 -05:00
• e15401ee0e Add runai-model-streamer into Python packages installed in Dockerfile and fix NotADirectoryError Docker regression (#22537) Alexis MacAskill 2026-04-14 16:25:41 -07:00
• 222eda1598 [Misc] Use cache_once for is_arch_support_pdl in sgl-kernel (#22725) Lianmin Zheng 2026-04-14 15:22:10 -07:00
• e83560562b Update CI Permissions (#22826) Jimmy Shong 2026-04-14 15:13:31 -07:00
• 8092431316 [serving] replace O(n²) stream_buffer string concat with integer offset (#22606) 3 people 2026-04-14 14:48:44 -07:00
• 36891ab514 Rename _alive_streaming_session_count; use _is_streaming helper (#22755) Liangsheng Yin 2026-04-14 13:26:03 -07:00
• 0cb7295698 Fix streaming session busy-check double-counting via active_pool_idxs (#22753) Liangsheng Yin 2026-04-14 13:11:06 -07:00
• b4616dcbf5 [BugFix] Fix EAGLE speculative decoding missing grammar-based finish … (#21723) mingyue300 2026-04-15 03:43:50 +08:00
• d2f479e544 [diffusion] chore: auto-enable best parallel setting if unspecified (#22763) MickandClaude Opus 4.6 2026-04-15 00:02:05 +08:00
• 3c0a6c6987 Add page_size and SWA coverage to unified radix cache bench test (#22815) Ke Bao 2026-04-14 23:58:05 +08:00
• 070c6a2489 [sgl] perf optimization for eplb (#21232) Bi Xue 2026-04-14 07:52:17 -07:00
• 9f9e0231bb Refactor unified radix cache UT into parameterized test suite (#22812) Ke Bao 2026-04-14 22:34:33 +08:00
• c5e95080d2 [diffusion] model: support Ltx 2.3 two stage ti2v (#22667) Mick 2026-04-14 22:10:08 +08:00
• 680bd4b429 [NPU] Modify the parameter name and optional values, and add the parameter restrictions. Modify some parameters supported type. (#22804) chx96642264 2026-04-14 21:34:07 +08:00
• 1588856e9b [NPU] qwen3next low latency best practice docs. (#22808) McZyWuandroot 2026-04-14 21:21:37 +08:00
• ddc7daaf89 [NPU] [DOC] Update NPU docs to match latest code (#22796) amote-i 2026-04-14 21:10:28 +08:00
• 454228e071 hicache storage backend mooncake support ascend hixl (#20016) lawtherWu 2026-04-14 20:51:06 +08:00
• 074c2a476d fix:[NPU]correct the full name of then Kimi model (#22799) loading66 2026-04-14 20:15:22 +08:00
• 68dfffaaa3 Offloading docs update (#22795) jianzhao-xuandJianzhao Xu 2026-04-14 20:03:29 +08:00
• 88253c39b0 [Docs] Fix formatting of tool-call-parser options (#22793) xdtbynd 2026-04-14 19:21:31 +08:00
• 368cdfbe2f [NPU] [DOC] Fix outdated descriptions in the NPU documentation (#22707) amote-i 2026-04-14 19:21:15 +08:00
• 6da3aba6a5 perf: optimize PCG inductor path for FP8 models (#21734) Jia GuoandClaude Opus 4.6 2026-04-14 02:51:27 -07:00
• 3cb3f7c018 fix: EPLB dispatch OOB when shared experts fusion enabled under DeepEP (#22525) xutizhou 2026-04-14 17:33:27 +08:00
• 6760c790bd [bugfix] avoid attention padding tokens computation in pcg (#17706) Jincong Chen 2026-04-14 16:08:23 +08:00
• eab045b2b7 [AMD] Add MiniMax-M2.7 accuracy and performance nightly tests (#22722) MichaelandHaiShaw 2026-04-14 00:30:11 -07:00
• d7ecab5113 [ROCm]fix(aiter): cast fp8 prefill output back to model dtype (#22626) xiaobochen-amdandkk 2026-04-14 15:25:09 +08:00
• f97c608caa [diffusion] quant: add FLUX.1-dev modelopt nvfp4 support (#22672) Xiaoyu Zhang 2026-04-14 15:00:59 +08:00
• 7c1bde2e38 [CI] Add optional image input to GB200 nightly workflow_dispatch (#22745) Sahithi Chigurupatiandishandhanani 2026-04-13 23:57:15 -07:00
• b10f852118 GLM-5/5.1 MXFP4 Checkpoint Inference Compatibility Fix (#22543) Colin ZandHAI 2026-04-13 23:56:48 -07:00
• 8fe9bbffb6 [CI] Reinstall flashinfer-jit-cache on CUDA version mismatch (#22741) Baizhou ZhangandClaude Opus 4.6 2026-04-13 23:04:23 -07:00
• d44eb16ac6 [AMD] Replace push trigger with scheduled runs and enable parallel stage execution (#22489) YC Yen-Ching Tsengandbingxche 2026-04-14 13:33:29 +08:00
• 657945c338 Replace all-reduce + dp_scatter with reduce_scatterv for DP attention (#22642) YAMY 2026-04-13 21:51:10 -07:00
• 520ce526b9 Restore Qwen3 rope config fallback (#22739) ishandhanani 2026-04-13 23:47:37 -05:00
• a9a2ae4a68 [Anthropic] Fix clock mismatch in received_time causing negative Prometheus metrics (#22247) Xuwei 2026-04-14 12:22:00 +08:00
• bc16130a17 ci: skip full rerun when sgl-kernel wheel already built (#22534) Jia GuoandClaude Opus 4.6 2026-04-13 20:32:55 -07:00