This website requires JavaScript.
4cc970290d
[CI] Fix duplicate job names that bypass branch protection (#22001 )
Liangsheng Yin
2026-04-02 23:59:35 -07:00
6b876a7710
[ROCM][RL] Shuffle Weight In-Place to Preserve Parameter Attributes (#21825 )
Yuzhen Zhou
2026-04-02 23:43:55 -07:00
75de479680
[Misc] Update CI permission (#22014 )
Baizhou Zhang
2026-04-02 23:37:05 -07:00
4d097047f2
[PD]: Add support for HiSparse to directly transfer the cache from Prefill to Decode DRAM. (#21591 )
2026-04-03 14:06:12 +08:00
5c082c307a
[Workflow] Fix kernel release jobs skipped on push events (#22011 )
Baizhou Zhang and Claude Opus 4.6
2026-04-02 23:03:13 -07:00
0a709cfe02
[Workflow] Avoid triggering nightly tests in kernel bump workflow (#22010 )
Baizhou Zhang
2026-04-02 22:40:33 -07:00
2c4fb88929
chore: bump sgl-kernel version to 0.4.1 (#21447 )
sglang-bot and sglang-bot
2026-04-02 22:31:59 -07:00
5bcbc9757c
[AMD] Resolve the performance degression when launch server with "--enable-aiter-allreduce-fusion" (#21947 )
kk and wunhuang
2026-04-03 13:10:24 +08:00
d1b7c3907d
[Parallel State Refactor 2/n] Unify code path of AMD deterministic all reduce (#20871 )
DarkSharpness
2026-04-03 12:33:17 +08:00
81efcc353a
[NPU] Optimized the wording in the npu docs (#21998 )
amote-i
2026-04-03 11:51:40 +08:00
efa7b2d5d3
Revert "[MUSA][9/N] Add FA3 attention backend support through MATE (MUSA AI Tensor Engine)" (#22002 )
Baizhou Zhang
2026-04-02 20:42:13 -07:00
5f0df1e2ad
[Bugfix] Fix incorrect dp-attention parallel info in bench_one_batch (#21519 )
lviy
2026-04-03 11:13:53 +08:00
69e89a1fcc
[VLM] Enable per-image MM splitting by default and remove MULTI_IMAGES modality (#21899 )
Yuhao Yang
2026-04-03 11:04:41 +08:00
8897ac58f0
[PP] qwen3 vl skip layer id for pp (#19135 )
narutolhy
2026-04-02 19:51:53 -07:00
991f3aa5b3
[Feature] NVFP4 Marlin fallback for non-Blackwell GPUs (SM75+) (#19652 )
Mook
2026-04-02 19:48:15 -07:00
2b5aed94f5
Remove maxItems=1 restriction when tool_choice is specified (#20208 )
Khoa Pham
2026-04-02 19:35:24 -07:00
0539c62bc1
[Diffusion][NPU] Add support for MOVA (#21633 )
Thomas and zhangshuai
2026-04-03 10:33:14 +08:00
1f97714f9b
[CI] Add timeouts to Slack upload urlopen and WebClient (#21903 )
Kangyan-Zhou and Claude Opus 4.6
2026-04-02 19:30:55 -07:00
89affff290
Skip broken AutoModel mapping entries when resolving Llava submodules (#21892 )
Xiaoyu Zhang
2026-04-03 09:04:26 +08:00
29d8e959d7
[CI] Remove stale Ascend suite entries from test/srt/run_suite.py (#21978 )
Baizhou Zhang and Claude Opus 4.6
2026-04-02 16:47:19 -07:00
34ddf135fd
[Feature] Stronger transformers modeling backend with TP, PP, MoE, VLMs, and torch compile (#19163 )
Adarsh Shirawalmath
2026-04-03 04:32:33 +05:30
939cf398a9
[MUSA][9/N] Add FA3 attention backend support through MATE (MUSA AI Tensor Engine) (#17985 )
ori and R0CKSTAR
2026-04-03 06:04:31 +08:00
566b4a4f1c
[4/n] Support gpt oss 20b lora (#21570 )
Ethan (Yusheng) Su
2026-04-02 12:57:38 -07:00
fe38410c3e
Remove logging for subprocess watchdog start (#21968 )
Lianmin Zheng
2026-04-02 11:30:33 -07:00
8732b2e9c6
[CI] [Tracing] Add ci for tracing and fix bugs (#21740 )
Feng Su
2026-04-03 01:50:50 +08:00
2278a321ca
[diffusion] chore: fix stage profiler for multi-stage denoising (#21955 )
Mick
2026-04-03 01:16:38 +08:00
df94cdcebb
[Parallel State Refactor 1/n] Remove stream of PyNCCL (#20866 )
DarkSharpness
2026-04-03 00:47:50 +08:00
b21db86e2f
[CI] Fix gpu deps import in cpu test (#21950 )
Ke Bao
2026-04-03 00:06:31 +08:00
083304ca44
[NPU] Support GLM-4.7-Flash on NPU (#21408 )
Todobe
2026-04-02 17:44:50 +08:00
9d9537fbd3
Migrate ngram corpus from torch cpp_extension to TVM FFI jit_kernel (#21920 )
Liangsheng Yin and DarkSharpness
2026-04-02 02:18:11 -07:00
b684b0b72f
Fix spec v2 + logprob when max_num_token is set (#20799 )
Qiaolin Yu
2026-04-02 01:55:16 -07:00
e55a35fbcd
test: add manual init test for mooncake transfer engine (#21842 )
foraxe and yunzhi
2026-04-02 16:01:10 +08:00
c7d03a6215
Revert "Rollback flashmla to older version [1/2]" (#21922 )
Baizhou Zhang
2026-04-02 00:27:02 -07:00
fbc1f92453
[DSA] Set trtllm kernels as nsa default for Blackwell (#21914 )
Baizhou Zhang
2026-04-02 00:22:27 -07:00
f30df723bf
scheduler: add prefill-only update in merge batch (#21840 )
Yilong Zhao
2026-04-01 23:33:06 -07:00
d24ea24e18
[NVIDIA] Enable fp8 flashinfer_trtllm_routed MoE for MiniMax-M2.5 (#20394 )
Trevor Morris
2026-04-01 23:02:06 -07:00
f25bf86065
Fix ngram doc for speculative_num_draft_tokens default (#21910 )
Liangsheng Yin
2026-04-01 22:18:24 -07:00
f836658077
[Spec][Ngram] 4/N: Remove max_match_window_size and min_match_window_size, matching all suffixes of the Trie (#21225 )
Khoa Pham
2026-04-01 22:09:46 -07:00
269589ad71
Return HTTP 400 for streaming validation errors (#21900 )
Liangsheng Yin
2026-04-01 21:58:12 -07:00
153359b4dd
Multi tool streaming fix (#20004 )
Khoa Pham
2026-04-01 21:53:05 -07:00
7a59e05dd1
[Kernel] Fuse temperature + softmax in sampling for decode speedup (#20501 )
Mook
2026-04-01 21:46:36 -07:00
afa14ffac9
Skip Go stdlib and NVIDIA tool CVEs in Trivy scan (#21905 )
Kangyan-Zhou and Claude Opus 4.6
2026-04-01 21:41:16 -07:00
cb0c2cbfdb
Enable multi-thread weight loading by default (#20289 )
Brayden Zhong
2026-04-02 00:27:20 -04:00
fae66b4050
Support PP key for file backend (#21901 )
Zhangheng
2026-04-02 12:23:58 +08:00
ed427e1299
Migrate all callers from /get_server_info to /server_info (#21463 )
David Cheung
2026-04-01 21:17:50 -07:00
24997fe42c
[diffusion] CI: add initial nvfp4 ci test for b200 (#21767 )
Prozac614 and Mick
2026-04-02 11:31:08 +08:00
648632b6c4
[CI] Remove crashing Kimi K2.5 EAGLE3/MTP variants, keep TP8 and TP8+DP8 (#21898 )
Kangyan-Zhou
2026-04-01 20:27:24 -07:00
9a7f198343
[CI] Increase multimodal server test timeout from 60 to 90 minutes (#21897 )
Kangyan-Zhou
2026-04-01 20:19:45 -07:00
875a615993
fix(ci): update est_time for 57 tests based on runtime analysis (#21896 )
Liangsheng Yin and Claude Opus 4.6
2026-04-01 20:16:13 -07:00
2ef12073f4
[VLM] Add VLM TP=4 per-commit CI test and improve MMMU eval prompt/parser (#21841 )
Yuhao Yang
2026-04-02 11:09:47 +08:00
7004df6094
chore: bump mooncake version to 0.3.10.post1 (#21844 )
Shangming Cai
2026-04-02 10:54:22 +08:00
0f6bedf6ed
fix pcg torch dynamo recompile in mxfp8 Triton path (#21888 )
Hanlin Bi and Hanlin Bi
2026-04-01 21:57:49 -04:00
8d9145d97e
Direct model loading from object storage with Runai Model Streamer (#17948 )
Noa Neria
2026-04-02 10:41:22 +09:00
ae3b207dfd
Allow /rerun-test to checkout fork PR branch for trusted users (#21890 )
Liangsheng Yin
2026-04-01 18:20:29 -07:00
51ad717089
[CI] Add Per-Tensor, Blockwise FP8 Tests on SM120 (#20717 )
Derek Yu and Brayden Zhong
2026-04-01 21:20:14 -04:00
83c3158014
[CI] Add Llama 3.1 8B Instruct FP4 CI test on SM120 (#20648 )
Derek Yu
2026-04-01 21:17:38 -04:00
6dd2f774de
[HiCache & PD]Fixed detailed cache hit breakdown in PD scenarios. (#21764 )
huangtingwei
2026-04-02 08:44:55 +08:00
9cb362f70e
[HiCache] fix: Clone host indices to avoid memory leak (#21624 )
shuwenn and Zhiqiang Xie
2026-04-02 08:42:07 +08:00
d7256eb69a
Unify GSM8K eval path to Chat API for regression CI readiness (#21667 )
Liangsheng Yin
2026-04-01 17:12:19 -07:00
1081a25983
revert: remove TTL-based hard pin from HiRadixCache (#21884 )
ishandhanani
2026-04-01 18:51:15 -05:00
1ac74e652e
[Misc] Fix comparator e2e tests: add polars dep + fix dp-attention test (#21804 )
Alison Shao and Alison Shao
2026-04-01 15:44:35 -07:00
70fc4ce3ef
Add merge prohibition policy during CI maintenance mode (#21882 )
Liangsheng Yin
2026-04-01 15:08:09 -07:00
821a8a99fb
[Disagg] GPU staging buffer with dynamic ring allocator for heterogeneous TP KV transfer (#19890 )
YAMY
2026-04-01 14:09:18 -07:00
5e12c4e08e
[DSA] Support trtllm sparse mla kernel for prefill batches (#21783 )
Baizhou Zhang
2026-04-01 13:55:05 -07:00
8950d129bd
[refactor] Clean up duplicate flashinfer trtllm moe code (#21233 )
Trevor Morris
2026-04-01 13:52:22 -07:00
0138708576
[Misc] Add network timeout to eval dataset downloads (#21873 )
Liangsheng Yin and Claude Opus 4.6
2026-04-01 13:16:14 -07:00
a19ef3a615
[FlashInver v0.6.7] Integrate flashinfer_trtllm mxfp8 gemm (#21576 )
Ziang Li
2026-04-01 12:55:06 -07:00
a1c725bdc5
fix: pre-init tokenizer_manager to avoid AttributeError in shutdown (#21824 )
shuwenn
2026-04-02 01:54:53 +08:00
ca3286d2d5
[diffusion] hardware: support FA3 attention backend on MUSA (attn backend, 14/N) (#18648 )
R0CKSTAR and Mick
2026-04-02 01:49:34 +08:00
6098c51bc2
fix(MiMo-V2-Flash): add mimo reasoning parser (#21414 )
shuwenn
2026-04-02 00:47:27 +08:00
c9f5d1d502
[Diffusion][NPU] add ring sp performance benchmark page in npu (#21811 )
yuefeng Wu
2026-04-01 23:53:10 +08:00
20f4193589
[Feature] JIT rmsnorm update (with claude) (#21834 )
DarkSharpness
2026-04-01 23:40:00 +08:00
4f5b55e379
[diffusion][CI]: Add individual component accuracy CI for diffusion models (#18709 )
Ratish P and Xiaoyu Zhang
2026-04-01 19:21:36 +05:30
e67b95d66b
[NPU]Add a full test pipeline on NPU, resolve issues in the NPU test architecture (#20751 )
Cherry_ming
2026-04-01 19:56:31 +08:00
ac039bd04e
Use CustomTestCase for TestSessionControl to enable CI retry (#21830 )
Liangsheng Yin
2026-04-01 04:26:11 -07:00
1aabe44b64
[VLM] remove AsyncMMDataProcessor wrapper (#21651 )
Yuhao Yang
2026-04-01 17:39:50 +08:00
80b1bc5f56
[NPU] update ascend docs (#21807 )
amote-i
2026-04-01 17:14:26 +08:00
7bba319f1e
[diffusion] fix: respect --prompt-path (#21756 )
Mick
2026-04-01 16:47:59 +08:00
95b881452e
Fix in-place mode in pause generation (#21705 )
wduan-hai
2026-04-01 01:36:28 -07:00
eec70286ec
[Bugfix] Fix effective_mamba_size over-allocation (#20858 )
yunkchen and Shangming Cai
2026-04-01 16:17:14 +08:00
7d2b856ce7
[Bug][VLM] Fix shared memory race condition in ShmPointerMMData broadcast for multi-GPU VLM serving (#21655 )
yudian0504
2026-04-01 16:15:14 +08:00
9eb75211b1
style refinement for hisparse (#21198 )
Zhiqiang Xie
2026-04-01 01:03:17 -07:00
57341b128f
glm_interleave for GLM-V (#21671 )
Yuxuan Zhang
2026-04-01 15:21:10 +08:00
835e19656f
Bug fix for llama eagle3 (#21397 )
khalilzhk
2026-04-01 15:01:53 +08:00
912494f596
[CI] Fix lint that was not applied in #21458 (#21818 )
Alex Nails
2026-03-31 23:58:12 -07:00
2861596fc6
[Bugfix] Fix PP tied embeddings weight loading for qwen3.5 4B dense model (#21347 )
Wenyao Gao
2026-03-31 23:51:03 -07:00
a188208e9a
[AMD] Optimize Qwen3-VL decode - fuse QK-norm + 3D mRoPE + KV cache write (#21458 )
2026-04-01 14:34:07 +08:00
71baa025be
Fix added tokens config with sensible filter (#17905 )
sbeurnier
2026-04-01 08:32:21 +02:00
87a2768269
VLM: change default mm-attention backend from triton_attn to fa4 (on blackwell) (#21595 )
Xinyuan Tong
2026-04-01 07:29:59 +01:00
72d3d8f4cf
[Feature Restoration] repetition_penalty is essential for GLM-V models (#21258 )
2026-04-01 14:29:49 +08:00
cffc95edf4
[3/n] lora moe - Support Qwen3-VL-30B-A3B-Instruct (#21469 )
Ethan (Yusheng) Su and Baizhou Zhang
2026-03-31 23:15:16 -07:00
ca3ba05a7a
chore: bump flashinfer version to 0.6.7 (#21422 )
2026-03-31 21:18:16 -07:00
03a87068ea
[KDA] Fuse scaled_dot_kkt + solve_tril + recompute_w_u for KDA (#21604 )
Yuan Luo and luoyuan.luo
2026-04-01 11:57:27 +08:00
e9b6cce237
[MPS] Fix Triton stub sub-module imports on Python 3.12+ (#21551 )
2026-04-01 08:56:01 +05:30
1b45d81e91
fix: only showing recent runners from ci failure analysis (#21015 )
Douglas Yang
2026-03-31 20:18:17 -07:00
ea6b22fb85
Fix CVEs in Docker image: pillow, linux-libc-dev, and broken sgl-model-gateway build (#21789 )
Kangyan-Zhou and Claude Opus 4.6
2026-03-31 20:07:15 -07:00
2488233ad5
[bugfix]GLM-4V model (#17122 )
KnightLTC
2026-04-01 10:37:40 +08:00
f9debd6514
[diffusion] CI: improve ci reliability (#21763 )
Mick
2026-04-01 10:06:57 +08:00
09907795e1
Add latency and throughput metrics to run_eval (#21793 )
Liangsheng Yin
2026-03-31 18:36:14 -07:00
8e84f846cc
[Diffusion] Add --uvicorn-access-log-exclude-prefixes to suppress noisy access logs (#20379 )
shuwenn
2026-04-01 09:31:44 +08:00