This website requires JavaScript.
503b898540
[Session] Count streaming session KV by owner in pool accounting (#42539 )
Liangsheng Yin
2026-10-04 22:15:57 -07:00
456d533e61
Revert "[AMD][ROCm] Keep cos_sin_cache fp32 on HIP for fused QSA indexer kernel (#41282 )" (#42541 )
Michael
2026-10-05 12:59:12 +08:00
1818db4e54
[CI] Drop the echoed command from /rerun-test result replies (#42543 )
Liangsheng Yin
2026-10-04 21:50:53 -07:00
b1bbd74f28
[XPU][CI] Disable test_xpu_graph until tc_piecewise is removed (#41634 ) (#42524 )
ashwini rathi
2026-10-05 08:39:26 +05:30
f385390be5
[Refactor] Simplify GDN, vision RoPE and MLA kernel dispatch (#42506 )
Xiaoyu Zhang
2026-10-05 09:37:53 +08:00
734cf3cf3b
[PD] Reload NIXL peer metadata after the peer is invalidated (#41691 )
2026-10-04 17:49:26 -07:00
f70e8c68fc
[Dsv4.1] Bounded replay with sparse mla path (#42273 )
Yuwei An
2026-10-04 17:19:14 -07:00
4b20bf4ea2
chore: update CI test est_time from recent run data
bot/update-est-time-20261005
sglang-bot
2026-10-05 00:00:30 +00:00
b792228b35
[ROCm] topk v2: split one long row across blocks, the CDNA cluster-path equivalent (#39931 )
Eric.Chin.AMD and Bingxu Chen
2026-10-05 07:54:37 +08:00
1f299e9f2f
[Refactor] Build layer stacks in order with append_stages (#42481 )
Cheng Wan
2026-10-04 16:40:24 -07:00
a99567ab74
[Fix] GigaChat 3.5: build the stage boundaries once (#42480 )
Cheng Wan
2026-10-04 16:39:34 -07:00
5c8fb9e938
[Fix] Offer the fused MoE finalize all-reduce only to a block that owes the sum (#42479 )
Cheng Wan
2026-10-04 16:39:06 -07:00
f07c1a398e
[Refactor] Build the Qwen4 experimental decoders from stage boundaries (#42478 )
Cheng Wan
2026-10-04 16:38:36 -07:00
92d60351e2
[Refactor] Build the Hunyuan V4 decoder from stage boundaries (#42477 )
Cheng Wan
2026-10-04 16:38:03 -07:00
f5ccc4d7d9
[Kernel] Use explicit reciprocal scaling for recurrent Q/K normalization (#42486 )
Lianmin Zheng and jiayisuse
2026-10-04 15:30:03 -07:00
d475b5a25c
[weight_cache] Coordinate daemon readiness across ranks (#38828 )
Tarang Khanna
2026-10-05 05:20:13 +07:00
caf96324fc
[Apple Silicon] Add an Apple Silicon (MPS) platform to the standard Torch model runner (#36780 )
Aditya Sharma
2026-10-05 03:33:40 +05:30
bab04cd791
fix(unified-memory): propagate unified-memory lazy checkpoint policy and handle unallocated session slots (#42072 )
SuperSong and Cheng Wan
2026-10-05 04:51:22 +09:00
fea3c088a1
[CI] Temporarily disable GB300 (#42463 )
Mohammad Miadh Angkad and Mohammad Angkad
2026-10-04 19:24:04 +00:00
aa0ce20201
[ROCm] GLM-5.2 decode path: decode-shaped MoE/MLA tiles, split speculative softmax, and bf16 GEMM routing (#41725 )
Zhang, Jiejing and kyle-256
2026-10-04 11:19:37 -07:00
7a719e9a65
[AMD] Fix RoPE cache dtype and diffusion CI failures (#42494 )
Xiaoyu Zhang
2026-10-05 00:36:26 +08:00
0343bc1483
[AMD][Docs] fix mori io and umbp playground (#42512 )
billishyahao
2026-10-05 00:28:11 +08:00
137c084000
[HiCache] refactor: retire in-flight storage prefetches through one helper (#41453 )
Shuwen Wang
2026-10-05 00:13:07 +08:00
affa261e3d
[diffusion] feat: enable cuda graphs for observation encoding and denoising steps for action models (#42171 )
WenhaoZhang
2026-10-04 22:22:08 +08:00
279d3f3888
[diffusion] feat: load ComfyUI/ai-toolkit fused gate_up LoRAs and diffusers metadata alpha for qwen-image-2.1 (#42174 )
WenhaoZhang
2026-10-04 22:21:13 +08:00
44fac689f5
[diffusion] optimization: decode json pixel lists to numpy at the entrypoint (#41895 )
WenhaoZhang
2026-10-04 22:20:20 +08:00
ab9d8f029c
[AMD][DI][CI] mi355x spur: skip RDMA ports that are not PORT_ACTIVE; dump driver log on failure (#41405 )
Michael
2026-10-04 21:23:54 +08:00
35f3c96ff4
[CI] Fix bootstrap sender registration test on main (#42487 )
Shuwen Wang
2026-10-04 18:45:31 +08:00
148da8e77d
[PD] Validate decode state layout once at registration (#42051 )
Shangming Cai and Claude Opus 5
2026-10-04 18:32:30 +08:00
bc5a055dd7
[diffusion] move kernel validation into launchers and simplify dispatch (#42391 )
Xiaoyu Zhang
2026-10-04 18:23:29 +08:00
29bcb74a89
[AMD][CI] Fix ROCm kernel wheel sources: drop eagle_utils, add DSV4 kernels (#42461 )
Michael
2026-10-04 17:54:28 +08:00
6cec8f98ce
[Bugfix] Fix unified-memory compaction gates and pending page reuse (#39982 )
SuperSong and Cheng Wan
2026-10-04 18:48:08 +09:00
1d02a36bb7
fix(inkling): translate unified-memory checkpoint destinations to physical slots (#38229 )
SuperSong and Cheng Wan
2026-10-04 18:45:45 +09:00
1e490772e5
[diffusion] fix: pin the JoyAI-Echo overlay's source revision (#42471 )
WenhaoZhang and Claude Opus 5.5
2026-10-04 17:19:09 +08:00
ba68306256
[diffusion] feat: let requests choose the video libx264 preset (#41797 )
2026-10-04 16:22:36 +08:00
fbce0d9478
[diffusion] fix: load serialized h3 INT8 in ComfyUI integrated mode (#42121 )
Tokha233 and WenhaoZhang
2026-10-04 16:08:56 +08:00
0bdd0cc8dc
[Deps] Upgrade the CUDA PyTorch stack to 2.14 (#38641 )
Mohammad Miadh Angkad and Mohammad Angkad
2026-10-04 07:48:02 +00:00
9631f6f2d0
[XPU] Enable fused QK-norm + RoPE for Qwen3-MoE (#40190 )
Kotthagattu Meher Sai
2026-10-04 13:15:18 +05:30
7eb628612c
[Fix] Publish the parallel config the CLIP attention test reads (#42460 )
Mohammad Miadh Angkad and Mohammad Angkad
2026-10-04 07:44:51 +00:00
55c5415c17
[CP] Support scattered interleave CP inputs with expert parallelism (#42041 )
Baizhou Zhang
2026-10-03 21:43:31 -10:00
937c0a6cfe
Fix unified HiCache physical transfers (#39479 )
2026-10-04 00:38:56 -07:00
6cc661f223
[AMD][DCP 2/N] Enable aiter asm ps for kimi k3 prefill (#34438 )
billishyahao
2026-10-04 14:34:45 +08:00
048b2a07f3
[Session] Release aborted streaming turns through the normal release path (#42330 )
Liangsheng Yin
2026-10-03 22:30:25 -07:00
1315d171a0
[Diffusion][CI] Add missing B200 Cirrascale runner E2E baselines (#42329 )
Mohammad Miadh Angkad and Mohammad Angkad
2026-10-04 04:10:32 +00:00
a3275295c4
[MoE] Use FlashInfer A2A for prefill instead of AG+RS (#39818 )
Brayden Zhong and Mohammad Angkad
2026-10-04 00:09:00 -04:00
c12bc635ca
[CI] Install only userspace libgdrapi for GDRCopy (#42233 )
Mohammad Miadh Angkad and Mohammad Angkad
2026-10-04 04:05:35 +00:00
0e6d7eba5f
[Fix] Count attention-DP ranks in one-batch bench running cap (#42331 )
Mohammad Miadh Angkad and Mohammad Angkad
2026-10-04 04:04:27 +00:00
fccc61c1c5
[CP] Support DP x TP x CP for attention with interleave CP strategy (#42038 )
Baizhou Zhang
2026-10-03 17:57:36 -10:00
f24db66729
[PD] Publish prefill DP rank whenever forced lookup is enabled (#41786 )
Aurick Qiao
2026-10-03 20:46:20 -07:00
169e6223ef
[PD] Drop the removed MooncakeKVSender args from test_prefill_complete_integration (#42454 )
metamergebot and metamergebot
2026-10-03 20:29:06 -07:00
4611ea365e
[HiCache]: Fix host lock ownership across radix splits (#38480 )
2026-10-04 05:16:04 +02:00
ff219e159e
[AMD] MiniMax-M3: fuse sparse QK norm, RoPE and cache writes with AITER (#35357 )
2026-10-03 19:49:53 -07:00
8afef58af4
docs: sync LMSYS SGLang blog cards (#41296 )
sglang-bot and sglang-bot
2026-10-03 19:26:05 -07:00
5606fb8592
[Fix] HiCache: carry registered Mamba slot side states (Qwen4-Exp PLE) through the host tier (#39862 )
2026-10-03 19:15:59 -07:00
f2385314b2
[AMD] Add DeepSeek-V4.1-Flash MI35x nightly accuracy test (#41476 )
Michael
2026-10-04 09:16:44 +08:00
663984b5a1
[AMD] Keep aiter's tuned GEMM for the DeepSeek-V4 compressors (#41931 )
Xinyi Song and kk
2026-10-03 18:11:37 -07:00
295a53c90e
[diffusion] optimization: fuse Flux3 rowwise FP8 quantization with triton (#41671 )
trebladev
2026-10-04 09:01:17 +08:00
71dc292b97
[AMD] Use AITER ASM prefill for MiniMax-M3 HD128 attention (#41707 )
2026-10-03 17:54:49 -07:00
4ab720e655
[HiCache] Fix cgroup page-cache accounting and the sizing fallback when cgroup discovery fails (#42420 )
2026-10-03 16:01:32 -07:00
4301be7bdb
Pick the default nccl_port below the kernel ephemeral range (#42405 )
metamergebot and merrymercy
2026-10-03 16:00:53 -07:00
10edb05e72
[Refactor] Let the remaining placement consumers read the parallel context (#42348 )
Cheng Wan
2026-10-03 14:59:32 -07:00
07064fd4a4
[sgl-router] Credit routed prompts before their KV events arrive (#42276 )
2026-10-03 14:57:49 -07:00
b2aa993cb8
[Fix] Skip a draft's decode recapture when it owns no decode graph (#42347 )
Cheng Wan
2026-10-03 14:54:12 -07:00
1a18de4b29
[Fix] Load a draft's tensor weight update from its deployment TP rank (#42346 )
Cheng Wan
2026-10-03 14:52:25 -07:00
061e712bab
chore: bump sgl-kernel version to 0.4.9 (#42427 )
sglang-bot and sglang-bot
2026-10-03 14:29:04 -07:00
3122b9c995
Update AOT kernels for Torch 2.14 (#42366 )
Mohammad Miadh Angkad and Mohammad Angkad
2026-10-03 21:26:34 +00:00
df53c978f1
[mem_cache] Run mamba models on UnifiedRadixCache when the radix cache is disabled (#42354 )
Liangsheng Yin
2026-10-03 14:11:13 -07:00
3e9a120add
[Fix] Make Hf3fsMockClient reads and writes thread-safe with pread/pwrite (#42425 )
Liangsheng Yin
2026-10-03 14:02:24 -07:00
8bf780dda5
feat(metrics): expose deferred decode KV release metrics (#41128 )
2026-10-04 04:52:21 +08:00
749a89abd6
[LoRA] Reorganize kernels and add CODEOWNERS (#42299 )
Yanbin Jiang
2026-10-03 13:43:40 -07:00
fd5e68f99e
[sgl-router] Repair KV-event sequence gaps from the engine's replay socket (#42275 )
2026-10-03 13:43:17 -07:00
aaf7ebe077
Advertise the KV-event replay endpoint in /server_info (#42274 )
2026-10-03 13:42:33 -07:00
3dc7f4b816
fix(cuda-graph): remove unnecessary CP batch alignment (#40543 )
AMD-yanfeiwang
2026-10-04 03:14:28 +08:00
e83c95b650
[AMD] Fix AITER weight slicing for CP decode TP (#39950 )
AMD-yanfeiwang
2026-10-04 03:09:08 +08:00
f327f92424
[AMD] Port CP V2 to the DeepSeek-V4 HIP backend (#34200 )
AMD-yanfeiwang
2026-10-04 03:08:17 +08:00
6acf2f4736
[PD] Keep ingesting requests while a prefill forward result is pending (#42035 )
2026-10-03 12:02:21 -07:00
b016ca406f
Support post-capture KV sizing for the unified hybrid-SWA pool (#41961 )
2026-10-03 09:54:20 -07:00
2bae12b9b3
[HiCache] Fix write-back SWA insert backups tripping the write-through pending-ack assert (#42264 )
2026-10-03 09:45:56 -07:00
5b5d721239
[PD] Add opt-in prefill-complete decode KV allocation (#40703 )
2026-10-03 09:03:09 -04:00
5916999afb
[DSv4.1] Fused c1/c2 compress for eager extend, faster c2 decode (#41660 )
2026-10-03 19:31:05 +08:00
d75d5b33f7
[DSv4.1] Faster fp4 index-K gather and combine_topk_swa_indices (#41658 )
2026-10-03 18:25:22 +08:00
379ec90fb2
[DSv4.1] Fold q_rope_store into fused_q_norm_rope (#41657 )
DarkSharpness and Claude Opus 5
2026-10-03 18:24:41 +08:00
7be5e3473c
[Refactor] Drop the reduction-skip mechanisms stage boundaries no longer use (#42312 )
Cheng Wan
2026-10-03 02:21:55 -07:00
71b04e02cf
[Refactor] Build the ZAYA1, IQuest-Q1 and Gemma 4 decoders from stage boundaries (#42311 )
Cheng Wan
2026-10-03 02:21:34 -07:00
f85c2c4923
[Fix] GigaChat 3.5: apply the sandwich norms to the complete sums (#42310 )
Cheng Wan
2026-10-03 02:21:10 -07:00
6faccf3e84
[Refactor] Build the GLM-4, GLM-Image and Granite MoE hybrid decoders from stage boundaries (#42309 )
Cheng Wan
2026-10-03 02:20:40 -07:00
e4554fd5e5
[Refactor] Build the Llama and Nemotron-NAS decoders from stage boundaries (#42308 )
Cheng Wan
2026-10-03 02:20:18 -07:00
740a4d5955
[Refactor] Build the Qwen2 decoders from stage boundaries (#42307 )
Cheng Wan
2026-10-03 02:19:56 -07:00
2c739b225b
[Fix] Jet-Nemotron build and Granite MoE hybrid final norm (#42306 )
Cheng Wan
2026-10-03 02:19:33 -07:00
b7fc516986
[Fix] EXAONE MoE under DP attention and DeepEP (#42305 )
Cheng Wan
2026-10-03 02:19:19 -07:00
2fa17a6467
[Refactor] Build the ERNIE 4.5 VL MoE and EXAONE MoE decoders from stage boundaries (#42304 )
Cheng Wan
2026-10-03 02:18:56 -07:00
f89ead192f
[Fix] EXAONE and ERNIE 4.5 VL MoE architecture, backend and PP issues (#42303 )
Cheng Wan
2026-10-03 02:18:33 -07:00
be20e7453c
[Fix] Step-3.5 DeepEP routed scaling and Sarvam shared expert under dense TP1 (#42302 )
Cheng Wan
2026-10-03 02:18:11 -07:00
b157d548bb
[Refactor] Let stage boundaries complete every stage-output sum (#42301 )
Cheng Wan
2026-10-03 02:17:48 -07:00
d5a8d76bf5
[Refactor] Drop TBO op methods that no strategy ever schedules (#42300 )
Cheng Wan
2026-10-03 02:17:26 -07:00
122b59152c
[mem_cache] Replace is_chunk_cache / is_tree_cache with supports_prefix_sharing (#42362 )
Liangsheng Yin
2026-10-03 02:07:55 -07:00
8293f9af54
[PD] Centralize drain-aware abort acknowledgements (#37077 )
Banruo Liu
2026-10-03 04:51:11 -04:00
1093c501df
[diffusion] optimization: route subBlock sparse attention in head chunks (#41985 )
saatwiknagpal
2026-10-03 00:54:54 -07:00
254aa41f78
[Fix] Guard DeepSeek NVFP4 shared-expert fusion for LoRA and FP4 backends (#42203 )
Yanbin Jiang
2026-10-03 00:37:57 -07:00
ce617d410b
[diffusion] fix: fix native FP8 format handling for FLUX 3 row-wise linears (#42255 )
Tokha233
2026-10-03 15:23:41 +08:00