Commit Graph

Select branches

Hide Pull Requests

dsv41-pd

dsv41-pd-visioncp

main

Mono Color

• a8c82c652e fix(dsv4): make pool configurator PP-aware (#25750) ybyang 2026-05-20 10:58:53 +08:00
• ca29c2b0e7 fix(dsv4): drop stale pp_size=1 guard for V4 PD disaggregation (#25771) ybyang 2026-05-20 10:57:23 +08:00
• d69e9fbfdc [diffusion] fix: honor config precisions for delight/paint (#22289) jy-song-hub 2026-05-19 19:38:48 -07:00
• 549ae16c6f [diffusion] fix: respect configured precision in Qwen layered path (#21980) jy-song-hubandjxp 2026-05-19 19:38:03 -07:00
• 3a9d9d5832 [diffusion] fix: fix Hunyuan3D-2 DiT checkpoint param mapping (#22729) jy-song-hub 2026-05-19 19:37:13 -07:00
• 65fe32379e [Intel GPU]Support fused_topk for XPU (#24641) gaopengff 2026-05-20 10:34:29 +08:00
• 0c8049d9ba Update CI permissions and CODEOWNERS (#25826) Baizhou Zhang 2026-05-19 18:19:21 -07:00
• 80fc524809 [diffusion] quant: update Wan2.2 modelOpt CI checkpoints (#25483) Xiaoyu Zhang 2026-05-20 09:05:39 +08:00
• 7f154ba449 drop output ids (#25774) Liangsheng Yin 2026-05-19 17:50:47 -07:00
• 425dffbde3 DeepSeek V4 MTP Support CP (#24934) Paiiiiandzengpai 2026-05-20 07:51:31 +08:00
• beaff00331 [NSA] Avoid repeated NSA MQA logits memory queries (#25299) YAMY 2026-05-19 16:04:13 -07:00
• 3ef832f885 pr-states: dispatch from pr-test* notify job (fix rerun status) (#25812) Liangsheng Yin 2026-05-19 15:57:16 -07:00
• b9d470f4a2 Support spec v2 for FlashMLA speculative decoding (#24640) 3 people 2026-05-20 06:23:17 +08:00
• 2f70902329 deflake priority below-threshold test (#25809) Liangsheng Yin 2026-05-19 15:19:01 -07:00
• b9c2bf717b [BugFix] Resolve adaptive speculative decoding conflicts for Qwen3.5 (hybrid GDN) (#23331) 3 people 2026-05-20 06:09:49 +08:00
• 16bcc4583e verify_done: wait not synchronize (#25465) Liangsheng Yin 2026-05-19 14:57:45 -07:00
• fab097d66d [Gemma4]: Fix FP8 Triton scale layout (#25286) Ratish P 2026-05-20 02:30:23 +05:30
• 8322fe09a7 fix(dsv4): upgrade forward metadata on main stream for large PP size (#25729) ybyang 2026-05-20 04:52:00 +08:00
• cd012ada58 [Fix] Fix extra uninstall of cutlass packages (#25756) Baizhou Zhang 2026-05-19 10:01:38 -07:00
• 4c0ce0345d Support Gemma4 Pipeline Parallelism (#25284) Yuan Luoandluoyuan.luo 2026-05-19 22:40:11 +08:00
• 3b62604cec [Diffusion] Support parallelism for GLM-Image (#25645) Makcum888e 2026-05-19 17:27:21 +03:00
• aad00b0ed8 Upgrade transformers to 5.8.1 (#25451) Xinyuan Tong 2026-05-19 15:20:30 +01:00
• 2bcb6d2f82 [Bug Fix] Align glm4_moe_nextn NPU MTP loading with qwen3 MTP (#25524) Yuxuan Zhang 2026-05-19 17:47:01 +04:00
• de3fc46e3d [NPU] [DOC] remove Qwen3-235B-A22B 2K+2K 100ms mixed mode benchmark (#25778) amote-i 2026-05-19 20:48:43 +08:00
• e0273dcd31 pr-test-extra: re-trigger on labeled event (#25732) Liangsheng Yin 2026-05-19 05:15:55 -07:00
• 0e4d1b49d3 [Codex] Remove stale DeepSeek V4 JIT kernels (#25764) Xiaoyu Zhang 2026-05-19 20:04:32 +08:00
• 45a85efc3a [Diffusion][NPU]Add attention backends for diffusion models for Ascend NPU (#23482) Arseniy MironovandNapkin-AI 2026-05-19 12:46:55 +03:00
• 58b5fe3e29 [Diffusion] [NPU] Fix HunyuanVideo crash on NPU (#25592) Thomas 2026-05-19 17:40:43 +08:00
• 5073c82a37 transformers v5 adapt HFRunner (#23922) jianzhao-xu 2026-05-19 17:07:38 +08:00
• 7e0818038a fix: fix deepseek v4 CP error (#25396) shiyu7 2026-05-19 17:04:21 +08:00
• 67fd005b97 [HiSparse & PD] Support hisparse memory pool host page > 1 (#23606) 3 people 2026-05-19 16:29:35 +08:00
• 1f7bf155c3 [NPU] [DOCS] Improved the usability of Ascend NPU documents (#25735) amote-i 2026-05-19 16:22:22 +08:00
• 78cb38ed5e [FlashInfer v0.6.11] [RL] Support FlashInfer per-token NVFP4 MoE (#22918) Ziang Li 2026-05-19 01:04:48 -07:00
• fbfddfd5c7 fix (jit kernel): elementwise activation C++ error (#25695) Kevin Li 2026-05-19 00:23:52 -07:00
• 79ea30d1f1 [Bug] Fix V4-Pro NaN on Blackwell by converting fp8_einsum input scale to ue8m0 (#25733) Yuhao Yang 2026-05-19 14:48:34 +08:00
• 7c3f614e23 [AMD] Bump amd/Kimi-K2.5-MXFP4 revision to align with shared-experts fusion (#25740) YC Yen-Ching Tseng 2026-05-19 13:47:05 +08:00
• 1d19721394 [gRPC] Native server: Rust crate (1/N) (#23506) Alex Nails 2026-05-18 22:31:38 -07:00
• 4c9f31b85e ✨ [diffusion][npu][quant] Add MXFP4 quantization support for Wan2.2 Diffusion on Ascend NPU (#22338) Junlin Wuandronnie_zheng 2026-05-19 12:46:52 +08:00
• 862d39e06c [Mamba] Fix extra_buffer overlap schedule races (#24954) Hanming Lu 2026-05-18 21:13:20 -07:00
• f0763859ed perf(mimo-v2-epd): enable GPU image preprocess and parallel video decode (#25588) Zhonghua Deng 2026-05-19 11:47:21 +08:00
• d8e66e54e5 fix: use triton_attn as default vision attention on B300 (SM103) (#25570) Yuhao Yang 2026-05-19 11:00:07 +08:00
• 31e324391b [Codex] Opt Mistral Large performace (#24611) Xiaoyu Zhang 2026-05-19 10:59:51 +08:00
• a7b3ced334 [diffusion] fix: fix LTX2 resident defaults and stage profiling (#25596) Mick 2026-05-19 10:41:28 +08:00
• 87c3c96bc8 [Bug][PD][NIXL] always send aux on is_last; only expects_state when truthy (#25699) ishandhananiandShangming Cai 2026-05-18 19:04:04 -07:00
• c2a212bfe2 [UnifiedTree] Support DeepSeek V4 host pool with multiple layouts. (#25282) huangtingweiandhzh0425 2026-05-19 09:36:00 +08:00
• b45b52ee8f Add spec_verify_calls_total metric for speculative decoding (#25689) Lianmin Zheng 2026-05-18 18:35:11 -07:00
• e4d81e48c9 Pull the max-prefix-len computation into its own helper and rename the matched-token argument (#25728) fzyzcjy 2026-05-19 09:27:06 +08:00
• 2424303dfb [codex] Optimize hidden-size 512 RMSNorm dispatch (#24710) Xiaoyu ZhangandCodex 2026-05-19 09:26:10 +08:00
• 170fe57cf0 Encapsulate the pending-flush bookkeeping in a small wrapper (#25727) fzyzcjy 2026-05-19 09:23:10 +08:00
• 5067da3eb6 Confine req-pool-idx assignment to the pool allocator (#25726) fzyzcjy 2026-05-19 09:22:51 +08:00
• 6bcbf4de35 Fix the misnamed request finish-check method to reflect its mutating semantics (#25725) fzyzcjy 2026-05-19 09:22:28 +08:00
• 2d2dff28da Return a mamba tracking entry from the cache lookup instead of mutating caller lists (#25724) fzyzcjy 2026-05-19 09:21:55 +08:00
• 3fd6a58e6c Inline the single-use split-prefill setup at its caller (#25722) fzyzcjy 2026-05-19 09:21:29 +08:00
• 1f3e5aa1e0 Publish elastic-EP active ranks from a dedicated step (#25721) fzyzcjy 2026-05-19 09:21:04 +08:00
• fa37b68653 Rename the request mid-chunk flag to describe what it actually tracks (#25720) fzyzcjy 2026-05-19 09:20:43 +08:00
• 32f1259c91 Confine max-prefix-len to where it is used and drop the leftover variable (#25719) fzyzcjy 2026-05-19 09:20:21 +08:00
• 1cba3ab467 Stop returning the unused prefix-computed flag from priority calc (#25718) fzyzcjy 2026-05-19 09:20:00 +08:00
• 2d868656d0 Move the retract-decode ratio estimation onto the new-token-ratio tracker (#25717) fzyzcjy 2026-05-19 09:19:42 +08:00
• 1a882c5c63 Pack scattered new-token-ratio state into a dedicated tracker (#25716) fzyzcjy 2026-05-19 09:19:20 +08:00
• 954b5c5846 Pack scattered scheduler IPC channel state into a dedicated container (#25714) fzyzcjy 2026-05-19 09:19:02 +08:00
• 07b4f262b7 Set up the idle sleeper outside of the IPC channel initialization (#25713) fzyzcjy 2026-05-19 09:18:36 +08:00
• 2d40f45193 Pack scattered request logprob state into a dedicated container (#25712) fzyzcjy 2026-05-19 09:18:19 +08:00
• fb7e49d4eb Expose can-run-cuda-graph as a regular property on the embedding result (#25711) fzyzcjy 2026-05-19 09:18:00 +08:00
• a740f8de33 Remove the dead hasattr fallback around the test-only crash counter (#25710) fzyzcjy 2026-05-19 09:17:40 +08:00
• 0e198f0f4f Refactor batch_result_processor into per-step prefill/decode helpers (#25709) fzyzcjy 2026-05-19 09:17:17 +08:00
• 7e7cb969e9 Route streaming output through the accumulator's payload method instead of an inline send (#25708) fzyzcjy 2026-05-19 09:16:42 +08:00
• da50e3d943 Log per-request time stats in a dedicated tail step (#25707) fzyzcjy 2026-05-19 09:16:22 +08:00
• b911fd1673 Route streaming-accept decisions through the accumulator instead of an inline gate (#25706) fzyzcjy 2026-05-19 09:15:45 +08:00
• d8f190dfba Pack scattered output-streamer state into a dedicated accumulator (#25705) fzyzcjy 2026-05-19 09:14:59 +08:00
• e8e55bb19b Split the request-reception loop into smaller phases (#25703) fzyzcjy 2026-05-19 09:12:47 +08:00
• d90bc65e30 [NPU] Fix TypeError in get_state_buf_infos when index_head_dim is None on MLA (#25383) Kurkur 2026-05-19 09:09:11 +08:00
• d028697d17 [NPU][Docs] Add Kimi-K2.5-W4A8 instance doc on NPU (#25269) Kurkur 2026-05-19 09:08:28 +08:00
• dbac464726 [Spec]: Make Triton standalone spec test deterministic (#25303) Ratish P 2026-05-19 06:06:10 +05:30
• b79e4b1e68 [Fix] Try to fix error caused by latest cutedsl packages (#25690) Baizhou Zhangandhnyls2002 2026-05-18 16:51:32 -07:00
• 314dedf7c6 Use SGLANG_CACHE_DIR env for gpu_p2p_access_cache path (#25686) 3 people 2026-05-18 16:00:54 -07:00
• 745abd6cc0 Add no_combine support to cutlass_moe_fp4 (#25688) Lianmin ZhengandHanming Lu 2026-05-18 15:59:55 -07:00
• 878e6b8886 [SP] Fix runtime_max_tokens_per_rank for sequence parallelism (#25685) 3 people 2026-05-18 15:59:33 -07:00
• 6f892047ec [misc] Throw error when single batch overlap is enabled on Hopper (#25509) Baizhou Zhang 2026-05-18 14:51:32 -07:00
• c904fdd20e ci: pr-states match renamed "PR Test Base" workflow_run (#25687) Liangsheng Yin 2026-05-18 14:42:29 -07:00
• 9e3bb9a307 [Spec] fold can_run_cuda_graph into EagleVerifyOutput; drop dead extend-after-decode check (#25566) Liangsheng Yin 2026-05-18 14:24:54 -07:00
• b7267e8fce [CI] Enable weight prefetch for 8-gpu-h200 basic tests (#25684) Cheng Wan 2026-05-18 14:12:45 -07:00
• 1f185c6ba8 Support draft extend cuda graph for tokenspeed_mla attention backend (#25489) Qiaolin Yu 2026-05-18 11:26:16 -07:00
• f5049709b3 fix(eagle3): drop +1 offset on aux layer ids when first id != 1 (#25454) Qiaolin Yu 2026-05-18 11:25:51 -07:00
• 86c6c77f2f [Bugfix] Fix missing group arg in get dp buffer (#25585) Shangming Cai 2026-05-19 02:08:10 +08:00
• b29e41e8b3 Respect user override for Gemma4 attention backend (#25547) Khoa PhamandClaude Opus 4.7 2026-05-18 10:46:25 -07:00
• f21fe6ad4d Fix PD disaggregation warmup: set request_name and improve error logging (#25542) Lianmin Zheng 2026-05-18 09:49:56 -07:00
• 3e2a109636 Add output_gate_type to Qwen3NextConfig and update models to utilize it (#25401) Gaoji Liu 2026-05-19 00:18:08 +08:00
• d96e593fd0 [Benchmark] Add SGLANG_SIMULATE_UNIFORM_EXPERTS for balanced expert routing with dummy weights (#25571) 3 people 2026-05-18 09:16:12 -07:00
• 866793c502 Amd/deepseek v4 rebase main 0509 (#24933) +313 people 2026-05-19 00:15:07 +08:00
• 110bbdcad7 [diffusion] fix: use dynamic LoRA for LTX2 original stage-two (#25591) Mick 2026-05-18 23:05:29 +08:00
• d1acd62d29 fix(disagg): unstuck decode aborts under prealloc pressure (#25561) ybyang 2026-05-18 22:57:22 +08:00
• 0ab427d0e1 [SMG] Add /v1/models fallback for model name discovery (#25293) grunerandAmit Gruner 2026-05-18 17:02:35 +03:00
• ba2ffcf156 Add DeepSeekV4 fused MoE Triton autotune support (#25569) 4 people 2026-05-18 21:35:32 +08:00
• f04c522534 [PD] Add conclude_state to fake KV backend (#25599) Shangming Cai 2026-05-18 19:56:42 +08:00
• be64d875ad Fix flush_cache AttributeError on is_stats_logging_rank (#25641) fzyzcjy 2026-05-18 19:23:48 +08:00
• 79b6749669 ci: pr-states no longer overwrites running run state when label is removed (#25586) Liangsheng Yin 2026-05-18 04:23:16 -07:00
• 8b94e1d0cf Delete the now-unused is_work_request from scheduler.py (#25639) fzyzcjy 2026-05-18 18:50:23 +08:00
• c54b34c007 Move module-level helpers out of scheduler.py (#25638) fzyzcjy 2026-05-18 18:45:38 +08:00
• 99ad2b0894 Move batch-result processing to SchedulerBatchResultProcessor and retire output_processor mixin (#25637) fzyzcjy 2026-05-18 18:45:01 +08:00
• 7d0b0b6991 Carve out SchedulerBatchResultProcessor for batch-result state (#25636) fzyzcjy 2026-05-18 18:44:41 +08:00