 Siyuan ChenandBJWang-ant
|
832ec39cc0
|
[Intra-node PD][DSV4] Pack all layers into one batch for INTRA_NODE_NVLINK path (#38984)
Co-authored-by: BJWang-ant <wangbingjia.wbj@ant-intl.com>
|
2026-09-15 21:08:33 +08:00 |
|
|
Yoray ZackandPo-Han Huang
|
1895cabfa8
|
Fix mooncake scale joiner groups (#38503)
Co-authored-by: Po-Han Huang (NVIDIA) <53919306+nvpohanh@users.noreply.github.com>
|
2026-09-15 19:07:25 +08:00 |
|
|
Yuxuan ZhangandXinyuan Tong
|
17ba2c2e7c
|
Support non-strict GLM47 tool calls with EBNF constraints (#38890)
Co-authored-by: Xinyuan Tong <xinyuantong.cs@gmail.com>
|
2026-09-15 16:41:48 +08:00 |
|
|
Qiaolin Yu
|
2c0a70960c
|
[qwen 3.8 next] reuse old cuda stream instead of endlessly creating streams (#39474)
|
2026-09-15 01:36:14 -07:00 |
|
|
|
4f52c9948b
|
keeping router GEMM in fp32 for deterministic inference (DeepSeek V3/V4) (#38176)
Co-authored-by: Brayden Zhong <b8zhong@uwaterloo.ca>
Co-authored-by: Brayden Zhong <brayden@radixark.ai>
|
2026-09-15 16:16:09 +08:00 |
|
|
Beihao ZhouandBrayden Zhong
|
37ebacb50f
|
[EPLB] Drop defensive getattr for ep_dispatch_algorithm (#31804)
Co-authored-by: Brayden Zhong <b8zhong@uwaterloo.ca>
|
2026-09-15 16:14:41 +08:00 |
|
|
Bi XueandLianmin Zheng
|
df254b0a11
|
[profiler] Label draft-runner steps DRAFT and target verify VERIFY in step spans (#38630)
Co-authored-by: Lianmin Zheng <lianminzheng@gmail.com>
|
2026-09-15 00:57:05 -07:00 |
|
|
iridiumine
|
8a20062652
|
[NPU][Bugfix] Disable pinned memory to fix DeepSeek-V2 DP-attention hang (#39423)
|
2026-09-15 14:59:31 +08:00 |
|
|
|
06992c92ad
|
[Fix] Aggregate all IPC weight update responses (#39534)
Signed-off-by: bojiang-li <327132355+bojiang-li@users.noreply.github.com>
Co-authored-by: hnyls2002 <lsyincs@gmail.com>
Co-authored-by: Liangsheng Yin <hnyls2002@gmail.com>
|
2026-09-14 23:25:46 -07:00 |
|
|
Liangsheng Yin
|
860fa83a9f
|
[CI] Wait for a killed test server's GPU memory before the next launch (#39545)
|
2026-09-14 23:24:37 -07:00 |
|
|
Cui Lily
|
e687b8d6af
|
[CPU] Implement fused QK Norm and RoPE kernels (#37748)
Signed-off-by: Cui, Lily <lily.cui@intel.com>
|
2026-09-15 14:17:57 +08:00 |
|
|
Khoa PhamandClaude Fable 5.1
|
2c37b90ad6
|
[Fix][DCP] Localize widened KV ids in MLA retraction CPU backup/restore (#39487)
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
|
2026-09-15 13:33:41 +08:00 |
|
|
|
ddd4600197
|
[Fix] HiCache startup ImportError on the pinned kernel wheel (#39516)
Co-authored-by: Mohammad Angkad <mohammad.angkad@radixark.ai>
Co-authored-by: Liangsheng Yin <hnyls2002@gmail.com>
Co-authored-by: hnyls2002 <lsyincs@gmail.com>
|
2026-09-14 22:31:19 -07:00 |
|
|
Shuwen Wang
|
5298d85218
|
fix: stop shadowing the DSpark shared-experts fusion guard (#39366)
|
2026-09-14 22:24:49 -07:00 |
|
|
ashwini rathiandClaude Opus 4.7
|
b510881157
|
[Fix][Qwen-VL] Normalize <image> sentinel on artifact fast path (#39278)
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
2026-09-15 13:19:59 +08:00 |
|
|
|
6c514ab025
|
Force reasoning mode for GLM-5.3 chat templates (#39227)
Co-authored-by: Brayden Zhong <b8zhong@uwaterloo.ca>
Co-authored-by: Mohammad Angkad <mohammad.angkad@radixark.ai>
Co-authored-by: Xinyuan Tong <xinyuantong.cs@gmail.com>
Co-authored-by: Xinyuan Tong <115166877+JustinTong0323@users.noreply.github.com>
|
2026-09-15 13:16:26 +08:00 |
|
|
Yanbin Jiang
|
ebd37705e4
|
[PD][LoRA] Gate decode admission on adapter slots (#39332)
|
2026-09-15 13:11:04 +08:00 |
|
|
ishandhanani
|
3f871a246c
|
feat(agent sessions): attribute stored KV cache blocks to sessions (#37482)
Signed-off-by: Ishan Dhanani <ishandhanani@gmail.com>
|
2026-09-14 21:34:49 -07:00 |
|
|
Byron HsuandByron Hsu
|
c9fbe5f655
|
[MM] Add flag to force Kimi image preprocessing onto CPU (#39148)
Co-authored-by: Byron Hsu <byron+per@periodiclabs.ai>
|
2026-09-14 21:19:39 -07:00 |
|
|
Aurick QiaoandAurick Qiao
|
5dde6e8f02
|
Fix MegaMoE buffer allocation and caching for effective SM budgets (#39223)
Co-authored-by: Aurick Qiao <6137920+aurickq@users.noreply.github.com>
|
2026-09-15 10:24:02 +08:00 |
|
|
Yuan Luoandluoyuan.luo
|
99060191e7
|
[KDA] Support ReplaySSM ring-write in the fused chain-verify kernel (#36821)
Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>
|
2026-09-15 10:22:41 +08:00 |
|
|
JoeandXiaoyu Zhang
|
a23fd557ed
|
[Kernel] Add OOT dispatch for clamp position (#38687)
Co-authored-by: Xiaoyu Zhang <1182563586@qq.com>
|
2026-09-15 10:22:21 +08:00 |
|
|
Kan Wu
|
07e1918924
|
[Rust] Use Dynamo native renderers when chat templates are missing (#38939)
|
2026-09-14 17:29:10 -07:00 |
|
|
Xingyu Liu
|
8874c51a96
|
[Benchmark] Add an opt-out for the token-capacity check (#39284)
|
2026-09-14 17:17:12 -07:00 |
|
|
raghothamandLianmin Zheng
|
c0b8725f5a
|
Fix /model_info serialization when a config value is a class (#39237)
Co-authored-by: Lianmin Zheng <lianminzheng@gmail.com>
|
2026-09-14 17:10:44 -07:00 |
|
|
Yang Liuandyangliu991
|
276663a79d
|
[model-loader] Split weight loading from postprocessing (#34981)
Co-authored-by: yangliu991 <yangliu991@fb.com>
|
2026-09-14 17:10:19 -07:00 |
|
|
Lianmin Zheng
|
a2b4e8888f
|
Allow CUDA VMM feature transport with the Rust frontend (#39347)
|
2026-09-14 17:09:49 -07:00 |
|
|
faceless void
|
6e755e4114
|
[Logging] Downgrade missing TokenizerManager request state log to warning (#36625)
Signed-off-by: syd520zy <529477025@qq.com>
|
2026-09-14 16:18:56 -07:00 |
|
|
AMD-yanfeiwang
|
0163f8ff74
|
[AMD] Fix registered HiCache host pointer aliases (#35233)
|
2026-09-14 15:49:05 -07:00 |
|
|
Baizhou Zhang
|
2fca6d69aa
|
bumping sgl-deep-gemm to 0.2.0 (#39371)
|
2026-09-14 15:30:31 -07:00 |
|
|
cctry
|
dad8c074e7
|
Scope prefetch cache state to the request attempt (#39318)
|
2026-09-14 15:03:03 -07:00 |
|
|
Qiaolin Yuandehuaa
|
d72e59508b
|
Reland fix(qsa): clamp the compress gather to the rows (#38346) (#39446)
Co-authored-by: ehuaa <ehuamail1@gmail.com>
|
2026-09-14 14:05:41 -07:00 |
|
|
paulzhang-tm
|
9128d57966
|
[Spec] Allow speculative workers to stage prefill shared reads (#38554)
|
2026-09-14 14:03:27 -07:00 |
|
|
Byron HsuandByron Hsu
|
5c2de3f355
|
[PD] Preserve the prefill rank during rebootstrap (#39357)
Co-authored-by: Byron Hsu <byronhsu@users.noreply.github.com>
|
2026-09-14 09:20:37 -07:00 |
|
|
pllimax
|
7465e42b7a
|
[NPU][CI] Add CANN 9.1.0 and Ascend a5 nightly suites (#38833)
|
2026-09-14 22:39:42 +08:00 |
|
|
pllimax
|
433c999dd0
|
[NPU][CI] Fix sglang.test.ascend import failure in multi-node e2e pods (#39403)
|
2026-09-14 22:23:47 +08:00 |
|
|
Kurt Shuster
|
f81fbc749a
|
[Fix] Merge adjacent KV-row frees so a mid-page split under DCP cannot double-free (#38941)
|
2026-09-14 21:44:32 +08:00 |
|
|
Aurick QiaoandAurick Qiao
|
2123aca87e
|
[Fix] Wait for PDL before reading DeepSeek V4 K cache locations (#38409)
Co-authored-by: Aurick Qiao <6137920+aurickq@users.noreply.github.com>
|
2026-09-14 21:04:28 +08:00 |
|
|
iridiumine
|
87db743021
|
[NPU] Fix device mismatch in SWA mask for DSpark verify graph capture (#39353)
|
2026-09-14 19:36:33 +08:00 |
|
|
Liangsheng Yin
|
66c7bc838e
|
[misc] Revert #38346, #33426, #39061 and #39219 (#39405)
|
2026-09-14 03:04:07 -07:00 |
|
|
amd-danli103
|
5aa9b8fb3e
|
[AMD][DSV4] feat: enable fp8 two-pool unified_kv on gfx950 (#37413)
|
2026-09-14 02:49:11 -07:00 |
|
|
Jacob0226andThomas Wang
|
5200508b0f
|
[AMD][gfx95] Fill the chunked-prefill compute budget exactly (#32888)
Co-authored-by: Thomas Wang <thomawan@amd.com>
|
2026-09-14 00:35:15 -07:00 |
|
|
|
3eeb7d37f9
|
[AMD] gfx950 assembly attention: length-aware split-KV for dynamic workload (#39172)
Co-authored-by: Zijie Chen <300606707+zijiecode@users.noreply.github.com>
Co-authored-by: jacky.cheng <yichiche@amd.com>
|
2026-09-13 23:48:03 -07:00 |
|
|
ZeyuanChen2000
|
a4781c9fe5
|
[NPU] Fix error due to missing parameter quant_linear passing (#37384)
|
2026-09-14 14:37:44 +08:00 |
|
|
Shuwen Wang
|
0d95a9c1ff
|
[HiCache] Keep the file backend temp file name within NAME_MAX (#38925)
|
2026-09-14 05:12:28 +00:00 |
|
|
Mohammad Miadh Angkadandmmangkad
|
2fd835b9c1
|
[Fix] Don't write conv state from the fused KDA verify kernel (#39219)
Co-authored-by: mmangkad <mohammad.angkad@radixark.ai>
|
2026-09-13 21:26:57 -07:00 |
|
|
Jan Bernlöhr
|
60f6f03409
|
Fix MUSA detection in compiled prefill path (#39061)
|
2026-09-13 21:23:51 -07:00 |
|
|
Kurt ShusterandBaizhou Zhang
|
f2111715cd
|
[fa] Make the FlashAttention backend extensible by subclasses (#33426)
Signed-off-by: Kurt Shuster <kurt@thinkingmachines.ai>
Co-authored-by: Baizhou Zhang <sobereddiezhang@gmail.com>
|
2026-09-13 21:21:25 -07:00 |
|
|
|
39e147443b
|
[Session] Fix image append positions and parent metadata (#39145)
Co-authored-by: Byron Hsu <byron+per@periodiclabs.ai>
Co-authored-by: Manik Singhal <manikvsinghal.pub@gmail.com>
|
2026-09-13 21:06:47 -07:00 |
|
|
Jackey Hua
|
5a132c061b
|
[Perf] trtllm_mla: reuse the fused fp8 KV/Q prepare on target verify (#39232)
|
2026-09-13 20:59:00 -07:00 |
|