This website requires JavaScript.
f4158719d1
[AMD] Let the diffusion AITer backend take grouped-query K/V (fix Cosmos3-Nano startup) (#34485 )
2026-08-19 00:17:40 -07:00
c0c87e0547
VLM: feed the packed qkv projection output to vision backends uncopied (#35336 )
Mick and Claude Opus 5
2026-08-19 14:39:52 +08:00
a5c96362b6
[diffusion] chore: reuse shared checkpoint quant metadata resolver (#35174 )
Mick
2026-08-19 13:36:55 +08:00
593b1a9b8a
[diffusion] optimization: reduce minimax h3 mps memory pressure (#33880 )
Mick
2026-08-19 13:35:01 +08:00
3391ab3712
[Constrained] Support MistralCommon tokenizers in the XGrammar backend (#35215 )
Alison Shao and Xinyuan Tong
2026-08-18 21:34:48 -07:00
a72d8d29d3
Fix HiCache PP sync test fixture (#35446 )
Shuwen Wang
2026-08-19 12:25:17 +08:00
c863760ae1
[Fix] DCP: advertise the logical KV-event block size (#35298 )
Jimmy Shong and Claude Fable 5
2026-08-18 20:55:11 -07:00
88f6074392
feat(api): add sglext_spec (#33518 )
Muqi Li and Codex
2026-08-19 11:54:18 +08:00
3b065a56b0
[HiCache] Batch PP write and load completion sync (#33473 )
luoroger37
2026-08-19 11:27:27 +08:00
1beb805356
[Perf] Restore the 16-token router GEMM threshold on SM10X (#34953 )
Mohammad Miadh Angkad and Mohammad Angkad
2026-08-19 10:53:52 +08:00
ee1f2e8dfd
[diffusion][Minimax H3]support subblock sparse attention on SM90 (#34680 )
HuangJi
2026-08-19 10:31:45 +08:00
eb085524c8
[diffusion] fix: make MiniMax-H3 AdaLN cache rebuild transactional (#34993 )
Dayuxiaoshui and Mick
2026-08-19 10:27:38 +08:00
977412ae61
[HiCache] Buffer-only mode for HiCache host memory layer (#34798 )
Zhiqiang Xie
2026-08-18 19:21:24 -07:00
4cef72faee
[diffusion] refactor: reuse srt qwen vision and text modules (#35006 )
Mick
2026-08-19 10:12:44 +08:00
58c5bee3ac
Fix DP attention on CPU (#12961 )
Chunyuan WU
2026-08-19 09:56:21 +08:00
5d12280ae7
Add fmha_v2 attention backend for SM90/120 (#23112 )
akhilg-nv and Yangmin Li
2026-08-18 18:41:03 -07:00
baa2251847
[diffusion] chore: make --vae-tiling honest, fix the decode oom advice, gate nvfp4 on blackwell (#35353 )
Mick
2026-08-19 09:27:54 +08:00
ef490853bb
quant: extract shared checkpoint quant metadata resolver (#35172 )
Mick
2026-08-19 08:26:41 +08:00
e73201e462
[diffusion] feat: support cache-dit, cfg gating, attention backend override as per-request param (#35339 )
Mick and Claude Fable 5
2026-08-19 08:24:03 +08:00
77fc5c128e
[perf] overlap page preprocessing, pack the vit, enable prefill CUDA graph for paddle-ocr (#35318 )
Mick and Claude Opus 5
2026-08-19 08:20:55 +08:00
64e404263e
[Doc] Fix TP and attention-TP group layout in initialize_model_parallel docstring (#34862 )
2026-08-19 08:15:03 +08:00
c14312a664
[Spec] DFlash2: local convolution + candidate selector (#35371 )
2026-08-18 17:07:28 -07:00
3a8f522f65
install sglang in virtual env instead of system path (#30612 )
Augusto Yao
2026-08-19 07:36:50 +08:00
cfc6dfb364
Apply latest DeepEP branch (#34923 )
Baizhou Zhang
2026-08-18 16:24:50 -07:00
37c09ff3d8
[Memory] Borrow CUDA graph pool storage for EAGLE sampling (#35375 )
cctry and cctry
2026-08-18 16:19:41 -07:00
7ebaa98f81
[Fix] Assert the page-aligned SWA evict floor on both PD decode prealloc paths (#35396 )
Liangsheng Yin
2026-08-18 15:48:55 -07:00
b814a7e812
[CI] Skip fast-fail for scheduled stages (#35392 )
Baizhou Zhang
2026-08-18 15:07:33 -07:00
87a09494fa
[Refactor] Share the page-aligned decode alloc lens between EAGLE and DFLASH (#35382 )
Liangsheng Yin
2026-08-18 14:39:41 -07:00
7b5410c999
Laguna: config-driven MoE router scoring (#35362 )
Joe Rowell
2026-08-18 22:42:36 +02:00
79dfef390b
[Spec] Page-align the DFLASH decode KV reservation (#35265 )
Liangsheng Yin
2026-08-18 13:36:04 -07:00
aa82229173
[Fix] Assert the page-aligned SWA evict floor at PD decode prealloc (#35286 )
Liangsheng Yin
2026-08-18 13:32:16 -07:00
955704544c
[Fix] Skip padded state slots in the chunked GDN kernel (#33431 )
Yuwei An and Claude Opus 5
2026-08-18 13:07:19 -07:00
307a90f6d3
Stop losing Kimi-K3 tool calls to reasoning, constraint conflicts, and truncation (#34881 )
2026-08-18 12:42:56 -07:00
8bb106cee9
Fix NIXL cleaner grouping for hybrid cache keys (#35130 )
Wei Yang and Wei Yang
2026-08-19 02:55:51 +08:00
526af15845
[Metrics] Discount queued prefill load by recent cache hits when waiting-queue matching is off (#35248 )
Hanming Lu
2026-08-18 11:27:21 -07:00
7dcaf11987
[Fix] Select custom all-reduce v2 by topology capability (#35061 )
Xingyu Liu and xingyuliu
2026-08-18 10:30:10 -07:00
480033def0
Refactor kv cache event mixin into a recorder (#35164 )
Ke Bao
2026-08-19 01:20:17 +08:00
83d7d45330
fix: preserve output logprobs without input logprobs (#34627 )
jain-ria
2026-08-18 10:07:13 -07:00
3f26febaff
[diffusion] fix: decouple encoder parallelism from the dit parallel layout (#34713 )
Mick and Claude Opus 5
2026-08-19 00:16:09 +08:00
9485c083bb
[NPU] Add mxfp4-w4a4 MOE Quantization Support for NPU (#30319 )
LinyuanLi
2026-08-19 00:06:05 +08:00
97dedd1ce9
[PD] Deferred decode-side KV release for aborts mid-transfer (#35049 )
Shangming Cai and Claude Opus 4.8
2026-08-18 23:56:33 +08:00
c60952d933
Exclude multimodal-gen NPU jobs from fast-fail cascade (#35238 )
hhhh1252023
2026-08-18 23:48:48 +08:00
499e90a125
[NPU CI] Reorganize test output/log directory structure with workflow context (#33685 )
pllimax
2026-08-18 23:46:00 +08:00
63d783bbe0
[diffusion] optimization: INT8 Linear + pluggable DiT attention backends for MiniMax-H3 on consumer-level GPUs (#34581 )
WenhaoZhang
2026-08-18 21:44:39 +08:00
0065fbfae1
[Scheduler] Cap prefill-delayer queue target by admission capacity (#35191 )
paulzhang-tm
2026-08-18 09:40:13 -04:00
94eef833fe
[diffusion] rl: support cosmos3 (#34197 )
Andy Ye
2026-08-18 05:42:27 -07:00
ae6945e112
[kernels] Reorganize ops/diffusion by operator domain behind a lazy facade (#35114 )
Xiaoyu Zhang and Claude Opus 5
2026-08-18 20:37:43 +08:00
7605529bdf
Add deepseek_v4_flash_w8a8_8p_in32k_out1k_50ms (#35162 )
HeYao and HeYao
2026-08-18 19:41:25 +08:00
f7101b0ae6
[AMD] MiniMax-M3 : Fuse QKV+index proj for block-fp8 (#32099 )
YC Yen-Ching Tseng
2026-08-18 12:01:01 +02:00
24d625698d
[AMD] feat(moe): fold padded-topk_ids fill into fused shared-experts append+remap (#31370 )
karverma-amd
2026-08-18 04:32:57 -05:00
70ee6b1714
[Rust Server] Add e2e latency metadata and fix Sarashina import (#35125 )
Lianmin Zheng
2026-08-18 02:08:07 -07:00
880ab72f34
test: extend NVFP4 Marlin tests to SM120 (#34327 )
vijay-kodamalla
2026-08-18 04:24:00 -04:00
fc0b95e7ba
Profiling Enhancements [2/3]: detailed execution step annotations (#24911 )
mohbasit
2026-08-18 04:09:07 -04:00
667389c50f
[diffusion] chore: filter transformer safetensors by index.json to drop duplicate shard variants (#35107 )
Emil Bogomolov and Emil Bogomolov
2026-08-18 01:05:05 -07:00
0111b29031
[Perf] Hoist DSv4 draft-extend SWA write locs; unify SWA graph buffer naming (#34890 )
Liangsheng Yin
2026-08-18 00:13:28 -07:00
a779a2a2a5
[Chore] Move version tag helper to release scripts (#35196 )
Lianmin Zheng
2026-08-18 00:11:12 -07:00
5b60b7c651
[DSV4] Turn on mhc post pre fusion by default (#35214 )
Baizhou Zhang
2026-08-18 00:03:37 -07:00
772018fe36
[diffusion] Per-section LoRA adapters on fused linear layers (#34933 )
Andy Ye
2026-08-17 23:40:45 -07:00
5eb117f6ca
[XPU] Fix decode graph runner is_current_stream_capturing on non-CUDA devices (#35050 )
ashwini rathi
2026-08-18 11:58:13 +05:30
27596abdc0
[AMD] Update amd k3 cookbook for PR#34580 (#35263 )
Thomas Wang
2026-08-18 14:19:36 +08:00
61600c9f39
[Diffusion][Refactor] Refactor and extract complex RoPE implementation to layers/rotary_embedding for MOVA DiT (#31453 )
Alexandr and ronnie_zheng
2026-08-18 09:14:35 +03:00
e6df23f3c2
refactor: rename chat response token IDs (#35225 )
Jiajun Li
2026-08-17 23:07:43 -07:00
0077f84d37
[mem_cache][8/N] refactor: move MambaPoolHost to pool_host.mamba (#31180 )
Shuwen Wang
2026-08-18 13:22:45 +08:00
ea27e3ddab
[AMD] Fix Quark Shared Experts Fusion Gate after load-time-override Removal (#35200 )
Colin Z
2026-08-17 22:10:01 -07:00
9401db3f29
[AMD] Scope the EAGLE greedy-verify TP broadcast to ROCm only (#35195 )
Zhiyao Jiang
2026-08-18 00:38:24 -04:00
8ea5229d42
[AMD] Add the Kimi-K3 MI35x perf benchmarks in nightly (#34985 )
2026-08-17 21:23:51 -07:00
d528192bf9
Skip inkling sheared bias under batch invariance (#35161 )
Ke Bao
2026-08-18 12:22:58 +08:00
d01812d89e
[AMD] Optimize KIMI-K3 with Triton MLA decode kernel by tuning the stage-1 geometry for gfx950 (#34580 )
amd-danli103 and Thomas Wang
2026-08-18 12:16:14 +08:00
53621818e4
[Docs] Enable PD disaggregation for DSV4 low-latency recipes (#35224 )
Baizhou Zhang
2026-08-17 20:07:27 -07:00
f44a130c5e
[DCP] Drop the prefill index-selection syncs by taking each rank's rows by stride (#35084 )
Khoa Pham and Claude Opus 5
2026-08-17 19:55:51 -07:00
fcdaaf8a5d
[Feature] Optimize TP LMHead with All-to-All (#32313 )
Siyuan Chen
2026-08-18 10:55:27 +08:00
d6c837489a
[XPU] Enable fused GDN QKV split Triton kernel on XPU (#30144 )
jiayisunx and Ma Mingfei
2026-08-18 10:42:58 +08:00
7f51e6bba0
[CPU] Explicitly import sgl_kernel in CPU kernel tests (#35119 )
Ma Mingfei
2026-08-18 10:20:28 +08:00
d55f1c28e2
[diffusion] feat: load quantized H3 text encoder checkpoints (#34986 )
Mick and Yiqi Yang
2026-08-18 09:10:54 +08:00
0ea262e6e5
[XPU] xpu kernel release workflow (#33679 )
Zaili Wang
2026-08-18 09:07:48 +08:00
c2c1c4d3eb
[CI] Move DSA PD+MTP+CP Layersplit test to basic B300 test suite (#35220 )
Baizhou Zhang
2026-08-17 17:44:31 -07:00
9ffc2856fb
docs: sync LMSYS SGLang blog cards (#35218 )
sglang-bot and sglang-bot
2026-08-17 17:30:28 -07:00
91144797c5
Update Qwen3.5 H200 FP8 for AgentX HiCache MTP (#35194 )
Faradawn Yang
2026-08-17 17:12:07 -07:00
c0b6474b43
[Spec] Reduce host-side overhead in ngram draft prep (#35207 )
Liangsheng Yin
2026-08-17 16:40:06 -07:00
b3c8f0d923
config: one control-plane log for the process (#35028 )
Cheng Wan
2026-08-17 16:19:00 -07:00
c70c7d72a8
config: the readback and the resolving view say what they are (#35027 )
Cheng Wan
2026-08-17 16:18:19 -07:00
cba3c5d5ac
config: the per-instance families read the bags (#35026 )
Cheng Wan
2026-08-17 16:17:53 -07:00
a97bc8db32
config: the DP/EP topology reads come from the parallel bag (#35025 )
Cheng Wan
2026-08-17 16:17:26 -07:00
d2bc697396
spec: size the speculative buffers from the bags, not the startup record (#35024 )
Cheng Wan
2026-08-17 16:16:56 -07:00
3d7ec00179
config: publish before a process reads configuration (#35023 )
Cheng Wan
2026-08-17 16:16:20 -07:00
2b278b4ac4
config: retire the multi-engine accommodation in the runtime context (#35022 )
Cheng Wan
2026-08-17 16:15:33 -07:00
bc312d185d
Clean deprecated DeepSeek V4 Environs (#34926 )
Baizhou Zhang
2026-08-17 16:07:00 -07:00
b42abbb1ba
[metrics] Fix prefill FLOPs estimate to count prefix and per-request causal pairs (#34316 )
Lennox Fu
2026-08-17 15:18:54 -07:00
b956e916ae
docs(cookbook): add Qwen3.8-27B DGX Spark configs (#35121 )
Jimmy Shong and Claude Opus 5
2026-08-17 15:04:17 -07:00
816ea65058
[AMD] Add Kimi-K3 8-GPU MI35x nightly accuracy CI (#32568 )
2026-08-17 14:48:29 -07:00
198a7b2fc9
[Misc] Clean up python/sglang package structure (#35062 )
Lianmin Zheng
2026-08-17 14:24:35 -07:00
770e7b47a2
[Spec] Support output logprobs with DSpark (#34478 )
2026-08-17 14:23:24 -07:00
032fe9c891
[Spec] Relay ngram accept tokens through the FutureMap (#35198 )
Liangsheng Yin
2026-08-17 14:21:07 -07:00
861eca8e25
docs: add NVFP4 quantization option to Kimi-K3 deploy panel (#35168 )
Yuhao Yang
2026-08-18 02:01:47 +08:00
2e7c85da68
[PD] Preserve decode KV across retraction in HiCache (#34801 )
cctry and cctry
2026-08-17 08:49:11 -07:00
af743371cc
Clean up environ.py: remove dead env vars, unify deprecation handling, move examples to a unit test (#35060 )
Lianmin Zheng
2026-08-17 06:53:34 -07:00
d97b796c16
[diffusion] chore: reuse SRT CLIP encoder blocks (#35004 )
Mick
2026-08-17 19:51:51 +08:00
6e8a4abb57
Add bit-exact class for MTP (#35143 )
Ke Bao
2026-08-17 19:51:39 +08:00
e9ad8102a2
[diffusion] chore: reuse SRT SigLIP in Pi0.5 (#34992 )
Mick
2026-08-17 19:33:36 +08:00
f33b83b4cc
[diffusion] fix: fix h3 swap peft SwiGLU lora_B halves when loading FFN Lora (#34940 )
WenhaoZhang
2026-08-17 18:47:31 +08:00