|
a6cf05817f
|
dsv4.1: remaining model and runtime integration (#38798)
Co-authored-by: BBuf <1182563586@qq.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Xiaoyu Zhang <xiaoyu.zhang@radixark.ai>
Co-authored-by: Yuwei An <ayw.sirius19@gmail.com>
Co-authored-by: Khoa Pham <khoa.pham@radixark.ai>
Co-authored-by: Yuhao Yang <47235274+yhyang201@users.noreply.github.com>
Co-authored-by: Zhichen Zeng <zczeng@uw.edu>
Co-authored-by: Ke Bao <ispobaoke@gmail.com>
|
2026-09-18 02:55:30 -07:00 |
|
|
Liangsheng Yin
|
1b200ffaaa
|
[Quant] Serve 32-wide-K ue8m0 block-FP8 linears through the FlashInfer MXFP8 GEMMs (#40039)
|
2026-09-18 02:51:12 -07:00 |
|
|
iridiumine
|
6c7c5e78de
|
[NPU] Run arch35 block-FP8 dense linears on the native MXFP8 GEMM (#39823)
|
2026-09-18 17:19:31 +08:00 |
|
|
iridiumine
|
7e6d5cbfac
|
[NPU] Gate DFlash replay metadata refresh behind spec_algorithm check (#39879)
|
2026-09-18 17:07:07 +08:00 |
|
|
iridiumine
|
d6090f92bf
|
[NPU] Fuse MXFP4 W4A8 MoE gmm1 + swiglu + requant into one kernel (#39881)
|
2026-09-18 16:50:49 +08:00 |
|
|
|
2dee23a876
|
[ROCm][diffusion] Enable fused qk norm and rope on ROCm (#35573)
Co-authored-by: jacky.cheng <yichiche@amd.com>
Co-authored-by: Claude <noreply@anthropic.com>
|
2026-09-18 16:40:32 +08:00 |
|
|
Shu Wang
|
1e8699fda3
|
[NVIDIA][comm] Merge EP+MoE-TP post-experts all-reduces into one _TP reduction (#32963)
|
2026-09-18 01:35:10 -07:00 |
|
|
zhaozx-cn
|
8ac39c66d8
|
[NPU] support kimi k3 on A5 and improve performance (#39589)
|
2026-09-18 16:33:54 +08:00 |
|
|
Rumit DesaiandXiaoyu Zhang
|
1fdd6c8921
|
[Runtime] Let out-of-tree platforms provide full graph backends (#37969)
Co-authored-by: Xiaoyu Zhang <1182563586@qq.com>
|
2026-09-18 15:42:46 +08:00 |
|
|
Ziang Li
|
c46bf5e990
|
[MoE] Disable FlashInfer fused finalize by default for numerical accuracy (#40105)
|
2026-09-18 00:15:57 -07:00 |
|
|
Shuwen Wang
|
0be8a0af0e
|
[DSV4] fix: keep the TileLang JIT cache under SGLANG_CACHE_DIR (#39364)
|
2026-09-18 00:06:35 -07:00 |
|
|
iridiumine
|
f86f60081d
|
[NPU] Adapt hicache for K3 hybrid models (#39415)
|
2026-09-18 14:48:20 +08:00 |
|
|
Jensen
|
3ce3b4969f
|
[NPU] Avoid repeated BF16 wo_a weight transposes in DeepSeek-V4 decode (#39919)
|
2026-09-18 09:10:43 +03:00 |
|
|
Bojiang LiandXiaoyu Zhang
|
3d1b9e7549
|
[Bugfix] Include SM121 in DeepGEMM packed-scale selection (#39482)
Signed-off-by: bojiang-li <327132355+bojiang-li@users.noreply.github.com>
Co-authored-by: Xiaoyu Zhang <1182563586@qq.com>
|
2026-09-18 13:57:14 +08:00 |
|
|
Hanming Luandhanminglu
|
6e1d9c1464
|
[profiler] Ignore PREBUILT batches in profile-by-stage (#40098)
Co-authored-by: hanminglu <hanminglu@fb.com>
|
2026-09-17 22:33:55 -07:00 |
|
|
Chi McIsaacandMick Qian
|
6215aecd51
|
[diffusion] feat: add metrics support (#19084)
Co-authored-by: Mick Qian <mickqian@users.noreply.github.com>
|
2026-09-18 13:22:03 +08:00 |
|
|
Yuwei An
|
65ef55e2a8
|
[Scheduler] Add shortest-prefill-first scheduling (#40024)
|
2026-09-17 21:18:58 -07:00 |
|
|
|
826d5170ae
|
[Bug] Guard FlashInfer CUTLASS MoE against 0-token inputs (#38780)
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Xiaoyu Zhang <1182563586@qq.com>
|
2026-09-18 11:03:27 +08:00 |
|
|
Zixin HuangandClaude Fable 5.1
|
f447bb7080
|
[Kernel] Coalesce the KDA CuTe DSL decode state transpose: ~3x faster, bit-identical (#39680)
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
|
2026-09-18 10:53:13 +08:00 |
|
|
|
db39b7f961
|
[Fix] Guard conditional top-logprob keys in the completions echo path (#34776)
Co-authored-by: James Liu <jamesl@modal.com>
Co-authored-by: hnyls2002 <lsyincs@gmail.com>
Co-authored-by: Liangsheng Yin <hnyls2002@gmail.com>
|
2026-09-17 19:47:26 -07:00 |
|
|
Nan Jiang
|
740f57a02c
|
[Spec] Fix CDF boundary handling in TreeSpeculativeSamplingTargetOnly (#35798)
|
2026-09-17 19:41:52 -07:00 |
|
|
William Arnoldandishandhanani
|
0214954f26
|
[gRPC] Stream engine state changes (#39915)
Signed-off-by: William Arnold <7565007+Aphoh@users.noreply.github.com>
Co-authored-by: ishandhanani <82981111+ishandhanani@users.noreply.github.com>
|
2026-09-17 19:17:25 -07:00 |
|
|
YanbingJiangandMa Mingfei
|
a407915c17
|
[CPU] Avoid prefill CP predicates during decode graph capture (#39690)
Co-authored-by: Ma Mingfei <mingfei.ma@intel.com>
|
2026-09-18 09:21:05 +08:00 |
|
|
Liangsheng Yin
|
f65c70bb7d
|
[Kernel] Move CUDA and ROCm speculative kernels to JIT (#40033)
|
2026-09-17 17:27:26 -07:00 |
|
|
Nan Jiang
|
20518d8518
|
[Scheduler] Align RadixCache no-insert cleanup with kv_len_to_handle (#35204)
|
2026-09-17 17:00:59 -07:00 |
|
|
Xiaoyu Zhang
|
7bc9152447
|
[Test] Consolidate kernel tests under plural kernels tree (#39966)
|
2026-09-18 07:37:48 +08:00 |
|
|
Liangsheng Yin
|
1f0c73e9bd
|
[DSV4] Generalize attention metadata, sparse prefill, and KV pool over compress ratios (#39921)
|
2026-09-17 15:55:19 -07:00 |
|
|
Ren YuzhouandRen Yuzhou
|
4f52a27563
|
[AMD] Fix DSV4 FP4 dequant path for AITER on ROCm (#35123)
Co-authored-by: Ren Yuzhou <yuzhouo7@users.noreply.github.com>
|
2026-09-17 13:51:22 -07:00 |
|
|
Thomas Wang
|
6c73368c32
|
[AMD][DSV4] Enable hicache on deepseek-v4 fp8 unified attn (#37778)
|
2026-09-17 11:29:23 -07:00 |
|
|
|
1f60ddef5d
|
[PD] Introduce runtime role switching between prefill and decode (#28403)
Signed-off-by: huanglong <huanglong@linux.alibaba.com>
Signed-off-by: inkcherry <mingzhi.liu@amd.com>
Co-authored-by: huanglong <huanglong@linux.alibaba.com>
Co-authored-by: Shangming Cai <csmthu@gmail.com>
Co-authored-by: Huang Long <121648372+LLLL114@users.noreply.github.com>
|
2026-09-18 01:45:12 +08:00 |
|
|
Byron HsuandByron Hsu
|
a98d921658
|
[DP Attn] Fix crash for no token all-gather case (#39899)
Co-authored-by: Byron Hsu <byron+per@periodiclabs.ai>
|
2026-09-17 09:21:09 -07:00 |
|
|
|
25c9f724d4
|
fix(multimodal): handle tensor images in exact-token preprocessing (#30368)
Signed-off-by: Zhuangcheng(Jesse) Gu <zcgu@connect.hku.hk>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: Mick <mickjagger19@icloud.com>
|
2026-09-17 09:20:21 -07:00 |
|
|
pllimax
|
7ccbf5fd04
|
[NPU][CI] Increase e2e multi-node test timeout to 210 minutes (#39952)
|
2026-09-17 20:50:33 +08:00 |
|
|
vstone-w
|
a9fb1c3238
|
dsv4(npu): support prefill context parallelism with interleave and zigzag (#39427)
|
2026-09-17 19:59:41 +08:00 |
|
|
jacky.cheng
|
11c35b8433
|
[AMD] Load fused shared experts for Qwen4-Exp and Qwen3.5 MTP (#38878)
|
2026-09-17 19:56:31 +08:00 |
|
|
Arseniy Mironov
|
a1b4ec02ae
|
[NPU][Diffusion] FA MXFP8 and modelslim w4a4f8 and w8a8f8 support for Wan2.2 and FLUX (#39438)
|
2026-09-17 13:13:05 +03:00 |
|
|
Shuwen Wang
|
6ca866ea29
|
[HiCache][Perf] fix: batch HiCache D2H submits per step for hybrid pools (#39050)
|
2026-09-17 17:53:58 +08:00 |
|
|
yl3469andShuwen Wang
|
1a90ae6727
|
Add Agentic-Aware Tail-Optimized LRU eviction to the unified radix cache (#34012)
Co-authored-by: Shuwen Wang <47200617+alphabetc1@users.noreply.github.com>
|
2026-09-17 17:52:59 +08:00 |
|
|
silencejade
|
575759d90a
|
[NPU] Support nccl backend for --remote-instance-weight-loader (#39413)
|
2026-09-17 17:28:15 +08:00 |
|
|
|
e970453b43
|
[NPU]Refactor weight processing and add NPUSwigluLimit activation (#38420)
Co-authored-by: AndyLi429 <AndyLi429@noreply.gitcode.com>
Co-authored-by: Even Zhou <even.y.zhou@outlook.com>
|
2026-09-17 16:40:32 +08:00 |
|
|
chuyehandchuyeh
|
2d08cc5ede
|
[AMD][Spec] Enable GDN ReplaySSM target-verify on ROCm (#38184)
Co-authored-by: chuyeh <298092489+chuyeh@users.noreply.github.com>
|
2026-09-17 01:29:14 -07:00 |
|
|
|
acfde25d34
|
Carry deferred attention operands and reuse multimodal shared memory (#39870)
Co-authored-by: fei-xx <135589532+fei-xx@users.noreply.github.com>
Co-authored-by: jmswen <jmswen@gmail.com>
|
2026-09-17 01:17:15 -07:00 |
|
|
Lianmin Zhengandjiayisuse
|
882577451e
|
Restrict SafeUnpickler standard-library globals (#39858)
Co-authored-by: jiayisuse <jiayisuse@fb.com>
|
2026-09-17 01:09:55 -07:00 |
|
|
chuyehandjacky.cheng
|
71ef869ece
|
[AMD][Bugfix] Fix vattn_asm HIP error 709 under CUDA graph capture on ROCm 10 (#39513)
Co-authored-by: jacky.cheng <yichiche@amd.com>
|
2026-09-17 01:06:04 -07:00 |
|
|
Teng MaandTeng Ma
|
15b256bdb0
|
[HiCache] fix: resolve Mooncake local_hostname per node for runtime attach (#29668)
Co-authored-by: Teng Ma <11641725+stmatengss@users.noreply.github.com>
|
2026-09-17 15:51:21 +08:00 |
|
|
Shijin Zhang
|
aebae58b8c
|
[Moe] Fix flashinfer_trtllm silently dropping swiglu_limit clamped SwiGLU activation (#39920)
|
2026-09-17 00:25:17 -07:00 |
|
|
MickandMick Qian
|
f3c0256771
|
[diffusion] chore: remove retired auto-residency workload helpers (#39884)
Co-authored-by: Mick Qian <mickqian@users.noreply.github.com>
|
2026-09-17 15:16:35 +08:00 |
|
|
faceless voidandronnie_zheng
|
44bd359082
|
[NPU][Diffusion] Optimize SenseNova-U1 batched generation (#39382)
Signed-off-by: syd520zy <529477025@qq.com>
Co-authored-by: ronnie_zheng <zl19940307@163.com>
|
2026-09-17 10:08:20 +03:00 |
|
|
Liangsheng YinandBBuf
|
3401b75240
|
dsv4.1: Engram module and request history support (#39666)
Co-authored-by: BBuf <1182563586@qq.com>
|
2026-09-16 23:49:47 -07:00 |
|
|
|
8ae4a39b50
|
Gate mamba extra-buffer predicates on uses_mamba_radix_cache (#37474)
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: Ke Bao <ispobaoke@gmail.com>
Co-authored-by: Sam Shleifer <sam@thinkingmachines.ai>
|
2026-09-17 14:39:12 +08:00 |
|