929230a6f0
Fix GLM-OCR MTP multimodal embeddings and positions ( #39088 )
...
Co-authored-by: Xinyuan Tong <115166877+JustinTong0323@users.noreply.github.com >
Co-authored-by: Xinyuan Tong <xinyuantong.cs@gmail.com >
2026-09-18 21:10:04 -07:00
Mind Lab
111b905bd1
fix(openai): recover logprobs token bytes from token_id (UTF-8 fragments) ( #38604 )
2026-09-18 20:30:04 -07:00
DefTruth and copilot-swe-agent[bot]
f1fbbd17bb
[diffusion] chore: update Cache-DiT to 1.5.1 for DMD Calibrator, SVDQuant DQ, etc ( #40104 )
...
Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com >
2026-09-19 11:14:33 +08:00
Vincent Liu
090263eff6
[VLM] avoid CUDA placement on non-CUDA platforms ( #38750 )
2026-09-19 11:13:22 +08:00
Alex and kpham-sgl
c3aa09b0db
[Kernel] Fuse hc_combine_norm for mid-size verify batches (9-96 rows) ( #40208 )
...
Co-authored-by: kpham-sgl <khoa.pham@radixark.ai >
2026-09-18 20:03:49 -07:00
hmalgewatta and Mick Qian
c475ac5eaf
[diffusion] feat: out of tree platform support ( #37547 )
...
Co-authored-by: Mick Qian <mickqian@users.noreply.github.com >
2026-09-19 09:35:58 +08:00
Xiaoyu Zhang
986959e3c4
[Refactor] Deduplicate kernel helpers and remove unused code ( #40197 )
2026-09-19 09:27:30 +08:00
Khoa Pham and Cursor
10b0bcfd18
[PD] Allow decode radix cache and HiCache L1/L2 with DCP ( #40263 )
...
Co-authored-by: Cursor <cursoragent@cursor.com >
2026-09-18 17:54:35 -07:00
Liangsheng Yin
d507accadc
[Test] Drop dead and strictly-subsumed CI test registrations ( #40264 )
2026-09-18 17:49:13 -07:00
Cheng Wan
fa7e83fd09
Name the two widths of the WORLD group ( #40070 )
2026-09-18 17:47:02 -07:00
Cheng Wan
81421b91e9
One read path for every parallel name ( #40069 )
2026-09-18 17:43:56 -07:00
Cheng Wan
afe71f4b9e
Read process groups through the runtime context ( #40068 )
2026-09-18 17:40:32 -07:00
5931fd60ee
Support unified memory page-envelope transfers in PD ( #39477 )
...
Co-authored-by: yhzhuang <yhzhuang@fb.com >
Co-authored-by: Lianmin Zheng <lianminzheng@gmail.com >
Co-authored-by: Yonghao Zhuang <yhzhuang@users.noreply.github.com >
Co-authored-by: Cheng Wan <cheng.wan@radixark.ai >
2026-09-18 17:39:50 -07:00
Cheng Wan
d0730a0e8b
Give the attention-DP width and rank one home ( #40067 )
2026-09-18 17:34:33 -07:00
Lifan Shen
8ea0ee300d
perf(sampling): avoid GPU syncs when applying custom logit processors ( #39234 )
2026-09-18 17:09:30 -07:00
Liangsheng Yin
b876213548
[Test] Add a ci-test-audit skill cataloging CI and test audit patterns ( #40257 )
2026-09-18 16:35:49 -07:00
Alison Shao and Xinyuan Tong
21e6c98ccb
Fix corrupted chat prompts on mistral_common tokenizers (tool_choice auto never fires) ( #39773 )
...
Co-authored-by: Xinyuan Tong <xinyuantong.cs@gmail.com >
2026-09-18 16:12:06 -07:00
2394b231c2
Fix Mistral3 retaining every vision-tower layer to read one ( #39185 )
...
Co-authored-by: Xinyuan Tong <xinyuantong.cs@gmail.com >
Co-authored-by: Xinyuan Tong <115166877+JustinTong0323@users.noreply.github.com >
2026-09-18 16:11:54 -07:00
fa521e2758
[MM] Copy placeholder ids to CUDA asynchronously ( #40010 )
...
Co-authored-by: metamergebot <metamergebot@users.noreply.github.com >
Co-authored-by: Jialin Ouyang <Jialin.Ouyang@gmail.com >
Co-authored-by: Lianmin Zheng <lianminzheng@gmail.com >
2026-09-18 16:10:37 -07:00
Zhiqiang Xie and cctry
ceb1d2e580
[PD] Enable optimistic prefill with buffer-only L3 write-through HiCache ( #40043 )
...
Co-authored-by: cctry <csycfl@gmail.com >
2026-09-18 16:04:15 -07:00
5e4b94b134
[MM] Skip VMM error gathers for text-only requests ( #40005 )
...
Co-authored-by: metamergebot <324680979+metamergebot@users.noreply.github.com >
Co-authored-by: Jialin Ouyang <Jialin.Ouyang@gmail.com >
Co-authored-by: Lianmin Zheng <lianminzheng@gmail.com >
2026-09-18 16:00:43 -07:00
hunhokim and Hun-ho Kim
803f0c93d2
Fix DSA partial DP-TP mode log to use derived attn_tp_size ( #39871 )
...
Co-authored-by: Hun-ho Kim <hunho.kim@samsung.com >
2026-09-18 15:54:45 -07:00
Jialin Ouyang
f3851486cb
[Perf] Fuse SWA page lookup and mapping clear ( #38948 )
2026-09-18 15:51:07 -07:00
81a199f56a
[HiCache] Read the in-flight buffer backup's node id from its snapshot in sanity_check ( #40013 )
...
Co-authored-by: Pranjal Shankhdhar <pranjalssh@meta.com >
Co-authored-by: Jialin Ouyang <Jialin.Ouyang@gmail.com >
2026-09-18 15:47:56 -07:00
0e5347db82
Support MXFP8 and deferred route weighting in DeepEP v2 ( #40030 )
...
Co-authored-by: metamergebot <324680979+metamergebot@users.noreply.github.com >
Co-authored-by: Xingyu Liu <38244988+charlotte12l@users.noreply.github.com >
Co-authored-by: pranjalssh <14260275+pranjalssh@users.noreply.github.com >
2026-09-18 15:40:38 -07:00
Liangsheng Yin
6cc9090d1f
[mem_cache] Release up to owned_kv_len on radix cache insert ( #40075 )
2026-09-18 15:37:52 -07:00
Zhiqiang Xie
a0534f8cca
[HiCache] Stop arming a prefetch retry for a too-short storage span ( #40042 )
2026-09-18 15:33:58 -07:00
Sam (Kesen Li)
d346b214fb
feat(kv-cache): support SM100 NVFP4 GenMHA and speculative decoding ( #36340 )
2026-09-18 14:50:46 -07:00
Lianmin Zheng and Jialin Ouyang
f5a1434700
[HiCache] Document transfer arguments ( #40239 )
...
Clarify the legacy host_indices argument and label Mamba test arguments,
including the current staging_tokens parameter. Executable code is unchanged.
Co-authored-by: Jialin Ouyang <Jialin.Ouyang@gmail.com >
2026-09-18 14:08:19 -07:00
amd-danli103
cd4dd81c22
[AMD][DSV4] fix: drop shadowing local get_exec import that breaks model startup on ROCm ( #40186 )
2026-09-18 13:28:38 -07:00
metamergebot and cctry
6a9c7001d3
[Logprob] Borrow graph-pool memory for input logprob logits construction ( #40007 )
...
Co-authored-by: cctry <csycfl@gmail.com >
2026-09-18 13:18:01 -07:00
Jialin Ouyang
da2f434951
[Spec] Add explicit prefill shared-read capability for plugins ( #39502 )
2026-09-18 13:11:01 -07:00
aed3fb1cdd
[Router] Log every request at one site; derive its outcome from the final status (3/3) ( #39465 )
...
Co-authored-by: Kangyan Zhou <kangyan.zhou@radixark.ai >
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com >
Co-authored-by: Shangming Cai <csmthu@gmail.com >
2026-09-19 03:11:00 +08:00
Lianmin Zheng and raghotham
248c202b46
Use runtime token widths for Triton speculative verification ( #39859 )
...
Co-authored-by: raghotham <853234+raghotham@users.noreply.github.com >
2026-09-18 11:07:55 -07:00
Lianmin Zheng
6bd1a0af1d
Add registration for external model configurations ( #39452 )
2026-09-18 10:06:24 -07:00
Byron Hsu
50a7de47d5
[Benchmark] Add agentic rollout simulator and offline explorer ( #40034 )
2026-09-18 10:06:13 -07:00
4e0b56c811
[Router] Treat an upstream 503/429 as backpressure, not a breaker fault (2/3) ( #39464 )
...
Co-authored-by: Kangyan Zhou <kangyan.zhou@radixark.ai >
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com >
2026-09-19 00:46:29 +08:00
Shuwen Wang
191172fa74
[Unified Tree] fix: exempt host-locked aux nodes from the sanity_check host-LRU check ( #39980 )
2026-09-18 15:29:29 +00:00
ChangLiu0709
45938a24ae
[AMD] GLM-5.2 MI355X MXFP4: bump image to 20260916, use HIP Top-K ( #40148 )
2026-09-18 23:21:58 +08:00
Xiaoyu Zhang
7714b182f2
[Bugfix] Fix top-1 MoE routing with non-unit scaling ( #40187 )
2026-09-18 22:53:13 +08:00
81363bf8cb
[kernel] Share the warp vectorized copy and enforce its alignment ( #36176 )
...
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com >
Co-authored-by: BBuf <1182563586@qq.com >
2026-09-18 22:40:54 +08:00
Mick and Mick Qian
9784d5f979
[diffusion] doc: sync CFG and tracing documentation ( #39883 )
...
Co-authored-by: Mick Qian <mickqian@users.noreply.github.com >
2026-09-18 18:03:40 +08:00
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
Cheng Wan
bbfcda48ce
[CI] Add a unified-memory rerun test group ( #40147 )
2026-09-18 01:59:07 -07:00
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