This website requires JavaScript.
ca36d88fa6
[ModelOpt] Fix broken Qwen3-235B-A22B-Instruct-2507-NVFP4 launch (#18189 )
Yi Zhong
2026-02-08 09:35:28 -05:00
43eecd8265
[diffusion] feat: support efficient sequence shard (#18161 )
wxy
2026-02-08 21:09:39 +08:00
f600965b0e
[CI] fix: notebook ci may not working (#18417 )
shuwenn
2026-02-08 14:26:35 +08:00
d71ccd8860
fix: sync server_args.kv_cache_dtype when detecting FP8 KV cache (#18394 )
Zack Yu
2026-02-07 22:10:59 -08:00
8e2e835c2f
[Fix] Fix backend selection after flashinfer version update (#18364 )
DarkSharpness
2026-02-08 11:20:41 +08:00
00248d85c7
[diffusion] platform: support WAN/FLUX/Qwen-Image/Qwen-Image-edit on Ascend (#13662 )
2026-02-08 05:45:30 +03:00
7b83659310
fix: fix NVFP4 Kimi-K2.5 weight mapping and exclude list (#18370 )
Mohammad Miadh Angkad
2026-02-08 10:23:48 +08:00
64950d8f97
[diffusion] feat: support saving videos directly on the server to avoid the overhead of tensor transfer (#18253 )
wxy
2026-02-07 22:08:42 +08:00
31d4cd2ffd
[diffusion] fix: respect dist_timeout option (#18386 )
Mick
2026-02-07 20:56:04 +08:00
fddef76619
[Doc] Fix outdated --fp4-gemm-backend documentation (#18350 )
Mohammad Miadh Angkad
2026-02-07 20:42:47 +08:00
d792aa7618
[diffusion] fix: remove unnecessary norm_type argument from GLM-Image dits (#18382 )
Hao Jin and Hao Jin
2026-02-07 04:35:12 -08:00
eb4cf1dfc4
[CI] Skip some flaky subtests for test_multi_lora_backend.py (#18408 )
Baizhou Zhang
2026-02-07 19:06:53 +08:00
c47c2f9466
[Doc] Update CUDA 13 install guide to install torch first (#18404 )
Mohammad Miadh Angkad and Baizhou Zhang
2026-02-07 18:04:37 +08:00
52401bec1d
chore: bump mooncake version to 0.3.9 (#18316 )
Shangming Cai
2026-02-07 17:30:01 +08:00
baec650462
[Diffusion] Apply fused_norm_scale_shift to LTX2/MOVA (#18257 )
Xiaoyu Zhang and yihanc
2026-02-07 17:28:42 +08:00
e834b85ab6
[NPU] update npu doc (#18344 )
Hexq0210
2026-02-07 16:38:05 +08:00
1552aab741
Support execute_shell_command for env var support (#18390 )
赵晨阳
2026-02-06 20:33:29 -08:00
4637970dfb
[Qwen3Next] Optimize fused_sigmoid_gating_delta_rule_update_kernel (#18271 )
hlu1
2026-02-06 19:59:42 -08:00
9fbec79906
Revert "[Build] Enable full kernel in aarch64 wheel" (#18385 )
Baizhou Zhang
2026-02-07 09:19:07 +08:00
bedade1ef0
Merge stage-c-test-large-4-gpu suites into partitioned suites (#18325 )
Alison Shao
2026-02-06 15:32:33 -08:00
f1ff697494
add hybrid model PD to NIXL connector (#16229 )
Neal Vaidya and ishandhanani
2026-02-06 15:05:36 -08:00
c850a8a41a
[Docs] Add Falcon H1, Hunyuan-Large, Qwen3-Omni support and update Diffusion usage (#17888 )
2026-02-07 02:47:51 +05:30
f2e0048d06
Add CI permission for Shunkangz, dongjiyingdjy, samuellees (#18377 )
Baizhou Zhang
2026-02-07 01:19:02 +08:00
e13b727e92
[diffusion] CI: update perf baseline (#17512 )
Prozac614
2026-02-07 00:28:44 +08:00
c6aa1863be
Add Nemotron 3 Nano tests (#18119 )
shaharmor98
2026-02-06 17:55:42 +02:00
79d409f210
[diffusion] fix: offload text encoder model in image encoding stage (#18317 )
xiaoye
2026-02-06 22:55:56 +08:00
3d68bd9d9b
add hicache jit test (#17847 )
Xuchun Shang
2026-02-06 16:54:33 +08:00
f798ab9775
[diffusion] fix: fix torch.compile graph break caused by torch._dynamo.disable (#18336 )
陈一涵
2026-02-06 14:48:09 +08:00
d0c39bc219
Fix cross-container HF download race condition in CI (#18328 )
Alison Shao
2026-02-05 21:01:41 -08:00
92b8bd6833
fix npu best practice (#18330 )
amote-i
2026-02-06 10:14:46 +08:00
aa390d2762
[Kernel] Migrate GPTQ-Marlin GEMM kernel to JIT (#18067 )
Linyu Wu
2026-02-06 08:31:42 +08:00
ef1d0ea885
[Doc] add a summary section for spec decode document (#18323 )
shuwenn
2026-02-06 05:34:31 +08:00
8b21dd4b77
[Doc] refine spec decode docs for SpecV2/STANDALONE/NGRAM (#18321 )
shuwenn
2026-02-06 04:12:33 +08:00
6a4b81e2d9
Refactor(qwen3-vl) optimize position encoding interpolation (#16781 )
2026-02-06 02:26:35 +08:00
d22163eb8c
Fix flaky test_frequency_penalty_reduces_word_repetition by using deterministic seeds (#18285 )
Alison Shao
2026-02-05 10:24:18 -08:00
498d8d0680
NixlKVManager optimizations (#17654 )
ovidiusm
2026-02-05 17:25:23 +01:00
b639779dd8
[diffusion] feat: allow T5's TP Group to reuse the transformer's SP Group (#17818 )
wxy and Mick
2026-02-06 00:12:19 +08:00
3f32a5831d
throw error if got adapter with added_tokens (#18046 )
Glen Liu
2026-02-05 10:55:43 -05:00
2eb4359ada
[Kernel] Add JIT apply_rope_with_cos_sin_cache_inplace (#18155 )
pansicheng
2026-02-05 21:49:37 +08:00
4aa03d91fd
[diffusion] fix: fix accuracy bug caused by #14717 (#18296 )
陈一涵
2026-02-05 20:36:18 +08:00
8f8c1724ae
docker: add patch to increase GPU deepep timeout (#18298 )
ishandhanani
2026-02-05 04:26:15 -06:00
afae4c7178
[PD] Minor code cleanup for mooncake backend (#18279 )
Shangming Cai
2026-02-05 17:38:09 +08:00
079fc8f3c5
[piecewise graph]: support MiniMax-M2 (#18217 )
zhangheng
2026-02-05 15:24:38 +08:00
3f1df322f9
[FIX] Always support TP > 4 for FP4 Gemm (#17300 )
danielafrimi
2026-02-05 09:10:26 +02:00
368936a62b
[XPU] Integrate MoE and minor improvements in XPU attention backend (#13561 )
Meng, Hengyu
2026-02-05 15:09:59 +08:00
dff3ba202a
[Diffusion] Support layerwise offload for mova (#18272 )
Xiaoyu Zhang
2026-02-05 13:16:07 +08:00
c910829708
Fix test_return_routed_experts to use response-level sglext (#18274 )
Alison Shao
2026-02-04 20:16:01 -08:00
e616d35847
Support Markdown/Notebook-Friendly Documentation Export for Downstream Integration(convert rat files to md files and save) (#18278 )
Kun Lin
2026-02-04 22:59:40 -05:00
de6a03260f
[docs] fix misspellings & typos (#18276 )
rinbaro
2026-02-04 19:35:29 -08:00
c8212b9fac
[PD] doc: Document SGLANG_MOONCAKE_CUSTOM_MEM_POOL and supported values (#18259 )
Teng Ma and Shangming Cai
2026-02-05 11:03:33 +08:00
f730c18679
[PD] improve kv offset calculation for MHA model with different tp size (#18163 )
Ch3ngY1 and Shangming Cai
2026-02-05 10:43:23 +08:00
f218234e4f
[diffusion] chore: prohibit Chinese characters usage (#18249 )
Mick
2026-02-05 09:22:26 +08:00
599c5f4922
fix kimi k2.5's moe gemm config init (#18064 )
yinghui
2026-02-05 07:59:01 +07:00
bdaf3de9b3
fix: add SGLANG_IS_IN_CI env var to release-docs workflow (#18225 )
2026-02-04 15:49:41 -08:00
c1d5cc3b24
[Bugfix] fix a obvious logic error (#18254 )
linhaifeng
2026-02-05 05:59:58 +08:00
efbf39583e
Add MoE fused config for Qwen3-Coder-Next-FP8 on H100 TP=2 (#18195 )
Mohammad Miadh Angkad
2026-02-05 05:36:35 +08:00
2e87c2bd5e
fix: fix MockModelRunner in attention tests (#18240 )
Zack Yu
2026-02-04 13:18:02 -08:00
6fd878b41d
[AMD] Add kimi mi35x nightly test, folder organization and several stability fixes (#17895 )
Michael
2026-02-04 12:03:57 -08:00
36a3e78af9
[diffusion] refactor: move model_stages into stages folder (#18248 )
Mick
2026-02-05 00:23:31 +08:00
3e7ecb78a6
model: support interns1-pro (#18145 )
RunningLeon and Ke Bao
2026-02-05 00:22:44 +08:00
a6f53cc5e3
entrypoint: support passing spaces_between_special_tokens per request (#17939 )
RunningLeon
2026-02-04 22:18:36 +08:00
4c403045ec
[diffusion] fix: fix the bug of redundant memory usage on GPU-0 (#18221 )
wxy
2026-02-04 21:25:23 +08:00
0c9a0adc53
[diffusion] chore: clean MOVA codes (#18107 )
Zhang Yiyang (SII)
2026-02-04 21:23:41 +08:00
760ae933bb
optimize get_topk_ragged by fusing get k and k_scale triton kernel (#16043 )
BingjiaWang and abing
2026-02-04 19:59:41 +08:00
315306d8a9
Make sure we always disable symm memory without dp padding (#18129 )
Nicolas Castet
2026-02-04 05:58:28 -06:00
a72f4f839c
Tiny fix for fp8 moe backend flashinfer_trtllm naming (#18243 )
Jincong Chen
2026-02-04 19:58:04 +08:00
ce02df8592
[diffusion] logging: downgrade default prompt log from info to debug (#17813 )
Evrard-Nil and Mick
2026-02-04 12:19:02 +01:00
2e9d0442e2
[diffusion] update code owner (#18247 )
Xiaoyu Zhang
2026-02-04 19:12:32 +08:00
84c09913eb
Moving _alloc_extend_naive out of npu allocator (#18200 )
Cheng Wan
2026-02-04 02:09:55 -08:00
be557cbc5f
[RadixTree][5/N Refactor]: Introduce pre and post-processing methods for key matching (#18147 )
zhangheng
2026-02-04 17:10:46 +08:00
d279520ba5
[DeepGemm] Add a flag for fast warmup (#18111 )
Baizhou Zhang
2026-02-04 14:12:13 +08:00
4739f2e8d5
[diffusion] kernel: gated residual layernorm scale shift and layernorm scale shift kernel fusion for Qwen-Image, WAN and HunyuanVideo (#14717 )
2026-02-04 13:46:20 +08:00
669a9bd180
Support Markdown/Notebook-Friendly Documentation Export for Downstream Integration (copy all markdown and rst files) (#18223 )
Kun Lin
2026-02-03 23:53:17 -05:00
b7c1dfc602
fix: bumping nightly whl version (#18212 )
Douglas Yang
2026-02-03 20:43:38 -08:00
37c33cc0aa
fuse qkvbfg linear into one gemm and f_b g_b into batched gemm. (#17801 )
strgrb
2026-02-04 11:41:26 +08:00
c1d529c196
Fix Session for multimodal and expose it through Engine (#18152 )
Aurick Qiao
2026-02-03 18:33:27 -08:00
1f72f66c6d
[Docs] fix readme typo (#18207 )
Qi Jia
2026-02-04 09:37:28 +08:00
da758ed601
[diffusion] fix: fix server cache-dit bug under continuous dynamic requests (#17140 )
wxy
2026-02-04 09:03:37 +08:00
ae004e15c9
fix: ensuring nightly whls are tagged with latest commit (#18204 )
Douglas Yang
2026-02-03 15:54:41 -08:00
793bf9fc06
Update weight rename check for Qwen3 Embeddings (#17535 )
satyamk7054
2026-02-03 13:55:11 -08:00
e867040fc6
add streaming parallel tool call test case (#18097 )
Hudson Xing
2026-02-03 12:46:01 -08:00
7de650c83c
[diffusion] hardware: support diffusion models on MTGPU (doc, 6/N) (#17346 )
R0CKSTAR
2026-02-04 04:44:57 +08:00
ec2461bc16
[diffusion] hardware: support diffusion models on MTGPU (multi-GPU, 5/N) (#17318 )
R0CKSTAR
2026-02-04 04:44:22 +08:00
acf724b036
[Diffusion] Only import sgl_kernel in custom op cuda path (SiluAndMul and RMSNorm) (#15592 )
R0CKSTAR
2026-02-04 04:42:58 +08:00
e166ca8758
[HiCache] feat: Add detailed cache hit breakdown for HiCache in sglext and Prometheus metrics (#17648 )
Vladislav Nosivskoy and ishandhanani
2026-02-03 22:45:35 +03:00
d48bbe3bed
[CI][NPU] Bugfix import sgl-kernel error (#18173 )
Even Zhou
2026-02-04 03:39:38 +08:00
495290aefd
enable ut test for xpu devices (#11712 )
2026-02-04 03:15:14 +08:00
0a6925639b
ci: improve docker for cu13 builds (#18194 )
ishandhanani
2026-02-03 13:09:38 -06:00
0db6fd4dbe
Revert broken sgl_kernel exclusion patterns in paths-filter (#18193 )
Kangyan-Zhou
2026-02-03 10:56:44 -08:00
820df545f2
fix: add cu13 dev container to our release (#18192 )
ishandhanani
2026-02-03 12:42:05 -06:00
99fab2ce67
[Bugfix] Fix Mistral Large 3 NVFP4 TRTLLM MoE (#18065 )
elvischenv
2026-02-03 20:32:49 +08:00
a45647bce1
[PD] feat: support mooncake intra-node nvlink kv transfer (#17866 )
2026-02-03 17:47:52 +08:00
cc69ac9e7a
Warmup before profiling prefill latency for dynamic chunk sizing (#17198 )
Xiaowei Wang and Shangming Cai
2026-02-03 17:45:23 +08:00
8e933e1914
AMD PD/D PR ci (#17183 )
2026-02-03 01:29:14 -06:00
25508d11c0
[Docker] Remove hardcoded America/Los_Angeles timezone, default to UTC (#18121 )
Mohammad Miadh Angkad
2026-02-03 15:22:15 +08:00
6f6b9c6e42
[Perf] Use safetensors load_file in multithread loader (#18124 )
Mohammad Miadh Angkad
2026-02-03 15:21:13 +08:00
7a9d9c79d1
[HiCache] fix: apply extra_backend_tag in Mooncake batch_exists (#17265 )
fatSheep
2026-02-03 14:54:56 +08:00
74f716dbd7
Gigachat 3 tool parser and tests (#14765 )
Viacheslav
2026-02-03 09:28:34 +03:00
4181290efd
[NVIDIA] Add --top-k argument to run_eval.py (#18025 )
Kaixi Hou and Claude Opus 4.5
2026-02-02 22:17:53 -08:00
fe57a887b1
[TestFix] use unit tests for LoRA overlap loading tests (#18140 )
Glen Liu
2026-02-03 01:06:50 -05:00