This website requires JavaScript.
3ad3f23ed5
[Comm] Drop the in-tree MNNVL CuTe DSL port in favor of FlashInfer 0.6.18 (#37206 )
Mohammad Miadh Angkad
2026-09-04 12:49:48 +08:00
e3305b3b87
Add code owner for sglang-simulator (#37922 )
Zhangheng
2026-09-04 12:20:48 +08:00
97d081ac76
[diffusion] chore: remove unreachable cosmos3 transfer encoding (#37805 )
Mick
2026-09-04 12:14:03 +08:00
ff1285cc28
[CP V1 Deprecation 2/5] Make strategy prefill CP canonical (#36223 )
Baizhou Zhang
2026-09-03 20:24:19 -07:00
59799a3687
[Simulator] Add high-fidelity CPU-based inference simulator (#33824 )
2026-09-04 11:12:11 +08:00
a5f07b1241
[CI] Build the Rust extensions for aarch64 too (#37820 )
Mohammad Miadh Angkad and Mohammad Angkad
2026-09-04 10:54:08 +08:00
8770c1db1f
[CPU][CI]: rename Xeon CPU CI suites to stage-*-intel (#37395 )
jundu and Ma Mingfei
2026-09-04 10:08:41 +08:00
c1b4d535d7
[CI] Fix lint (#37895 )
Mohammad Miadh Angkad and Mohammad Angkad
2026-09-04 09:37:50 +08:00
5e81d462c3
[diffusion] fix: only use fused qk_norm on nvidia gpu (#33994 )
Aleksi Vesanto
2026-09-04 04:25:08 +03:00
6c63d678d5
[diffusion] webui: fix minimax h3 webui inference settings (#36320 )
Li Jinliang
2026-09-04 09:24:14 +08:00
94eb15eb6c
[diffusion] quant: support fp8 mixed precision for cosmos3 (#36380 )
Kedi Wu and Kedi Wu
2026-09-03 18:23:38 -07:00
667bc043dc
[CI] remove MINIMAX_H3_HF_TOKEN (#37687 )
Yihao Wang
2026-09-03 18:22:32 -07:00
2a0602c7ac
[diffusion] optimization: unfused w2 bias on SM12.x (cuBLAS 16x16 kernel mis-dispatch) for minimax-h3 vae decoder: (#37835 )
Mick
2026-09-04 09:17:23 +08:00
795dd7abce
[diffusion] feat: compose third-party component bundles safely (#37816 )
Mick
2026-09-04 09:15:05 +08:00
516cfbd362
Fix UNO test adapter subdirectory resolution (#37872 )
eigen and Yingyi Huang
2026-09-03 18:07:09 -07:00
9ed2721c6d
[NPU] fix extend_seq_lens_cpu shape in eager mode (#36843 )
heziiop
2026-09-04 09:03:30 +08:00
6147a54ddf
[PD] Bound transfer engine init with SGLANG_DISAGGREGATION_ENGINE_INIT_TIMEOUT (#37874 )
Liangsheng Yin
2026-09-03 17:16:55 -07:00
66d60433c1
state_capturer: pin the exact host-cache size via mmap + cudaHostRegister (#37285 )
Yueming Yuan
2026-09-03 16:20:59 -07:00
8b0501399e
[CI] Improve Lark CI cards: structured layout, PDT timestamps, slow-only queue digest (#37884 )
Liangsheng Yin
2026-09-03 16:18:56 -07:00
b44496389c
[HiCache] Count hit allocations and in-flight backups in the buffer pipeline idle check (#37883 )
Zhiqiang Xie
2026-09-03 16:08:50 -07:00
a480f388b2
[HiCache] L3 storage prefetch lifecycle metrics and cross-tier attribution fixes (#37503 )
Zhiqiang Xie
2026-09-03 16:00:04 -07:00
0610a6539d
[CI] Add Lark notifications for CUDA CI status, runner health, and queue time (#37881 )
Liangsheng Yin
2026-09-03 15:46:33 -07:00
4dc9cda5f9
[PD] Gate deferred decode KV release on backend capability (#37454 )
Yonghao Zhuang and yhzhuang
2026-09-03 15:42:21 -07:00
fd70325c10
[CI] Add dspark + dsv4 e2e test (#37665 )
Baizhou Zhang
2026-09-03 14:26:25 -07:00
4372b8efa7
[1/N] Quantization Refactor: remove dead code and dedup the FP4 marlin helpers (#37552 )
Mohammad Miadh Angkad
2026-09-04 05:22:31 +08:00
0e5414fd2f
[Test] Allow top-k cutoff ties in test_sampling_mask_matches_topk_logprobs (#37873 )
Liangsheng Yin
2026-09-03 14:13:26 -07:00
05dbe64dff
Fix buffer-mode idle tracking and VLM memory sizing (#37567 )
Lianmin Zheng
2026-09-03 13:56:12 -07:00
2a980cbf10
[mem_cache] Require page-aligned starts in free_segment and drop the boundary trim (#37729 )
Liangsheng Yin
2026-09-03 13:28:33 -07:00
3ffacf949b
[Docs] [BugFix] Sync --tool-call-parser and --reasoning-parser lists with the code (#37788 )
Martin Hickey
2026-09-03 21:17:33 +01:00
d0c95f6c91
[HiCache] buffer mode: anchor-lock staged prefetches by default (#37464 )
Zhiqiang Xie
2026-09-03 12:08:22 -07:00
68978f8d52
[Scheduler] Count the parked chunked-prefill request in the busy mem check (#37502 )
Zhiqiang Xie
2026-09-03 12:08:17 -07:00
2da5802bfa
[Cookbook] DeepSeek-V4 DGX Spark: v2 image + Flash Official NVFP4 and Flash Vision FP4 cells (#37737 )
Jimmy Shong and Claude Fable 5.1
2026-09-03 11:43:01 -07:00
abed680320
[Unified Cache][5/N]: Integrate external linker mode end to end (#37381 )
Zhangheng and 晟海
2026-09-04 02:02:58 +08:00
619ab2bcce
Fix block-scale swizzling device placement (#37849 )
Lianmin Zheng
2026-09-03 10:47:04 -07:00
23ab10a63e
Support speculative decoding with unified SWA memory (#36403 )
2026-09-03 10:44:57 -07:00
33a22b1b08
[Cache] Forward fast prefix matching capability (#37844 )
cctry
2026-09-03 10:33:30 -07:00
54cadad151
[Router] Add composable scoring and eligibility policies (#37731 )
Vincent Gao and inkcherry
2026-09-04 00:01:41 +08:00
392841f47c
[Bugfix] Support K2 Horizon MoE without MoVA (#37825 )
Yash Akhauri and BBuf
2026-09-03 08:53:16 -07:00
bf71035d39
[diffusion] MiniMax-H3: tiered AdaLN plan cache (pinned-host tier + per-plan LRU) (#37266 )
triple-mu and mickqian
2026-09-03 22:09:21 +08:00
4e37882a93
[Diffusion][minimax-h3] Add SM120 support for SubBlock sparse attention (#37332 )
Quanli Li and 全力
2026-09-03 22:05:22 +08:00
3239baef25
[CI][NPU] Fix kimi_k2_6 16p in64k perf test and dsv4-flash testcases (#37760 )
pllimax
2026-09-03 21:57:14 +08:00
9cb38a3d57
[diffusion] feat: filter duplicate precision variants across custom loaders (#37616 )
dujifeng and mickqian
2026-09-03 21:54:11 +08:00
f5bed255c0
[diffusion] feat: support key masks on USPAttention's replicated-prefix path (#36735 )
Yihao Wang and Claude Fable 5
2026-09-03 05:13:33 -07:00
14444c6a04
[diffusion] feat: add maybe_record_function profiler spans for request phases (#35922 )
Yihao Wang and Claude Fable 5
2026-09-03 05:12:54 -07:00
2bb25dc18b
[Speculative Decoding] Add native UNO serving support (#37667 )
2026-09-03 05:08:41 -07:00
354ed6d66b
[NPU] [DOC] Refresh supported models and features on NPU (#37799 )
amote-i
2026-09-03 19:53:39 +08:00
dd091f43cd
[AMD] Update kimi-k3 amd cookbook 0903 (#37781 )
kk and wunhuang
2026-09-03 18:44:39 +08:00
a11dba1a01
[Feature] Unified memory: support decode context parallelism for the trtllm_mla family (#37693 )
Cheng Wan
2026-09-03 03:37:12 -07:00
f59a4840c5
[AMD][CI] Correct MI355X Slurm exclude node (#37779 )
YC Yen-Ching Tseng and Chen
2026-09-03 12:32:44 +02:00
429ac2d82c
[AMD] Fix DSv4 draft extend taking the target compression path during prefill (#37713 )
Duyi-Wang
2026-09-03 17:46:49 +08:00
27b7a2dc3b
[Kimi K3] Rework skipped-think fix as opt-in force_nonempty_content with streaming coverage (#34187 )
2026-09-03 17:37:28 +08:00
a6001478f4
[AMD] Perf Kimi-K3 MoE optimization (#33838 )
kk and wunhuang
2026-09-03 17:28:28 +08:00
397aeca376
fix(benchmark): support Glm4MoeLite in fused MoE tuner (#37623 )
Xiaoyu Zhang
2026-09-03 17:21:42 +08:00
7ed29eba80
[AMD] Fix FP4 indexer OOR (#37660 )
Xinyi Song and Thomas Wang
2026-09-03 01:49:46 -07:00
e59a576f03
fix test/manual/test_forward_split_prefill.py UT due to many refactors and design changes (#36617 )
Polisetty V R K Jyothendra Varma
2026-09-03 14:18:55 +05:30
3bac084d4e
[Model] Add native IFM K2 Horizon serving support (#37654 )
2026-09-03 01:39:43 -07:00
02d9b3060a
[Docs] Update K2 Horizon MoE model names (#37723 )
Yash Akhauri
2026-09-02 23:32:44 -07:00
1fb85053e7
[AMD][Diffusion] Migrate FlyDSL fused norm kernels to the v0.3.0 stable API (#36349 )
YC Yen-Ching Tseng and Phil Li
2026-09-03 08:03:08 +02:00
030d7e7e9b
[ROCm] Define the DSA head-gate graph helpers on HIP (#37118 )
xiaobochen-amd and Zhang, Jiejing
2026-09-03 13:55:45 +08:00
4b0edb25f2
[CPU] Update base image to Ubuntu 26.04 (#35313 )
Zaili Wang
2026-09-03 12:42:14 +08:00
54da74e83e
[Fix] Broadcast PP dynamic-chunk profiling failures so every rank disables together (#37675 )
Liangsheng Yin
2026-09-02 21:09:04 -07:00
98ef7d8ae6
docs: add K2 Horizon cookbook recipes and H200 results (#37655 )
Yash Akhauri
2026-09-02 20:52:05 -07:00
49e6e81830
[CI] Accept t64-suffixed apt packages in the install skip check (#37689 )
Liangsheng Yin
2026-09-02 20:51:47 -07:00
81b0e4985a
Modify KUBE_JOB_NAME to fix the problem of the string being too long (#37572 )
hhhh1252023
2026-09-03 11:43:36 +08:00
cf3173aeb9
[Perf] Walk the radix tree by offset instead of re-slicing token storage (ported from #36507 ) (#37324 )
Khoa Pham and Claude Opus 5
2026-09-02 20:13:03 -07:00
57c26a84e0
[chore] Add .git-blame-ignore-revs for the black -> ruff-format reformat (#37210 ) (#37695 )
Alex Nails and Claude Opus 5
2026-09-02 20:01:48 -07:00
4229088a48
feat(kernels): generalize persistent CuTe JIT cache (#33911 )
James Liu
2026-09-02 19:59:04 -07:00
28262c20df
[CI][RFC] Replace black-jupyter with ruff-format (#37210 )
Alex Nails and Alison Shao
2026-09-02 19:46:08 -07:00
2641e427be
Xpu/weekly simple model enablement 2026 08 30 (#37193 )
2026-09-02 18:35:59 -07:00
a522c8a4b6
[misc] Extract PP dynamic chunk sizing into a DynamicChunkSizer scheduler component (#37674 )
Liangsheng Yin
2026-09-02 18:35:43 -07:00
0dd66def7c
[chore] harden checkpoint quantization metadata parsing (#36922 )
Mick
2026-09-03 09:27:22 +08:00
fbf909b460
[Fix] Alpha-channel images and tool-result media ordering (port of #36507 ) (#37320 )
2026-09-02 18:23:06 -07:00
6e41f1ad29
[Fix] Preserve FP32 in SM107 MXFP8 fallback (#37489 )
Lee Nau and Yangmin Li
2026-09-02 20:55:14 -04:00
80e8302d03
Add SGLANG_CRASH_ON_JIT_COMPILE to forbid on-the-fly JIT compilation (#36615 )
Shiyan Deng and Lianmin Zheng
2026-09-02 17:26:25 -07:00
87d60a2229
Improve CUDA graph and speculative execution output handling (#37329 )
2026-09-02 17:25:27 -07:00
db1eb48651
[CI] Graceful teardown for the PD and HiSparse server fixtures (#37485 )
Alison Shao
2026-09-02 17:22:18 -07:00
ff04a00d73
Reduce tokenizer overhead and offload CUDA VMM publication (#37330 )
2026-09-02 17:21:08 -07:00
f15748d965
[bench] Support real-traffic replay with early-stop-aware steady-state metrics in bench_one_batch_server (#37469 )
Oguz Ulgen
2026-09-02 17:20:25 -07:00
5c46ce37f5
[Fix] Apply the attention-CP broadcast result in PP dynamic-chunk profiling (#37669 )
Liangsheng Yin
2026-09-02 17:20:07 -07:00
3421d4375b
[Docs] GLM-5.3-Flash cookbook: drop stale EP caveat, add B300/H100/B200 FP8 speed data (#37576 )
Xinyuan Tong
2026-09-03 08:16:23 +08:00
046cdaabaa
[Sampling] Capture masks from sampler support (#36630 )
Byron Hsu and ByronHsu
2026-09-02 17:13:28 -07:00
5ddca6819e
Fix unified SWA: size a non-owner's v2p by the id space it must address (#37560 )
Cheng Wan and Claude Opus 5
2026-09-02 16:56:18 -07:00
5a1275a519
Converge the two SWA predicates, and stop conditioning the capture sink on the pool (#37550 )
Cheng Wan and Claude Opus 5
2026-09-02 16:55:46 -07:00
d9848b9ecd
Build the unified read stream directly, without the page-table rectangle (#37512 )
Cheng Wan and Claude Opus 5
2026-09-02 16:55:16 -07:00
18d5ffb42a
Size the unified read-table grid from bs, and fuse the allocator's tombstone scatters (#37511 )
Cheng Wan and Claude Opus 5
2026-09-02 16:54:26 -07:00
c05f8ae830
[PD] Optimize paged allocator free-list release (#37146 )
2026-09-03 07:51:37 +08:00
4bc34117f1
[CI] Install lmms-eval from PyPI, drop human-eval install, add clone token fallback (#37672 )
Liangsheng Yin
2026-09-02 16:23:10 -07:00
a6da3a4922
[CI] Re-enable GB300 jobs (#37522 )
Alison Shao
2026-09-02 16:15:40 -07:00
718bd39fe0
[Fix] DP attention: correct the decode->extend prefix off-by-one (#37505 )
Mohammad Miadh Angkad
2026-09-03 07:14:16 +08:00
3fa6b86504
[Spec] Publish the final multi-layer EAGLE shared-read event (#36752 )
paulzhang-tm and Qiaolin-Yu
2026-09-02 18:57:50 -04:00
19c30dff56
[SM120] DeepSeek-V4: DeepGEMM paged-MQA indexer +FP4 MoE+ page-split (#29927 )
eeecho and Baizhou Zhang
2026-09-03 06:49:59 +08:00
982aa8acfc
[Bugfix] Load Qwen3.5 MTP embedding under PP (#37471 )
YAMY
2026-09-02 15:19:40 -07:00
3c9cea8f10
[EAGLE] Prune draft-extend logits to selected rows (#35546 )
YAMY
2026-09-02 15:10:08 -07:00
fe45af1e6f
perf(gdn): select ReplaySSM verify loop unrolling by shape (#36970 )
YAMY
2026-09-02 15:07:49 -07:00
9c70d22721
Update GLM-5.2 NVFP4 B200/B300 for AgentX HiCache (#35368 )
Faradawn Yang
2026-09-02 14:39:09 -07:00
3a855b050a
fix(disagg): poll receivers during decode preallocation (#37483 )
cctry
2026-09-02 14:26:40 -07:00
ad6e830858
[Bugfix] Key CUDA graph dedup signatures on kernel function identity (#37657 )
cctry
2026-09-02 14:24:31 -07:00
19c7679e9e
[mem_cache] Make free_swa sync-free on page_size == 1 (#36723 )
Liangsheng Yin
2026-09-02 14:18:22 -07:00
acea43079f
Fix native MoE handling of noncontiguous top-k IDs (#36407 )
Joe and BBuf
2026-09-02 17:16:19 -04:00
2d799c28f4
[CI] Authenticate and retry git clones in install scripts (#37647 )
Liangsheng Yin
2026-09-02 13:21:26 -07:00