 ZhenchengLin
|
6376fda7a9
|
[Bugfix][Rust Frontend] Use server-reported prompt_tokens in vllm-bench (#59694)
|
2026-10-28 09:22:50 +08:00 |
|
|
|
458334f00f
|
[MRV2][Spec Decode] Close DFlash/DSpark and adaptive-verification JIT warmup gaps (#63819)
Signed-off-by: mgoin <mgoin64@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
|
2026-10-28 01:15:00 +00:00 |
|
|
Matthew BonanniandClaude Opus 5.5
|
ebaa812475
|
[Bugfix][HiSparse] Pad staged prefill KV to avoid a trtllm-gen out-of-bounds read (#61659)
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
|
2026-10-27 21:00:46 -04:00 |
|
|
Matthew BonanniandClaude Opus 5.5
|
66f1d0fa16
|
[Bugfix][HiSparse] Read resident prefill rows by request state slot (#62739)
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
|
2026-10-27 20:57:29 -04:00 |
|
|
|
0106c8abce
|
[CI] Shard FP8 MoE kernels and L4 Humming eval automatically (#61831)
Signed-off-by: Kevin Luu <51931015+khluu@users.noreply.github.com>
Signed-off-by: Thang Nguyen <thangnguyenvn647@gmail.com>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
|
2026-10-27 17:54:04 -07:00 |
|
|
Woosuk KwonandCodex
|
a91b939bfb
|
[Bugfix][Bench] Support Muse balanced routing (#63872)
Signed-off-by: Woosuk Kwon <woosuk@inferact.ai>
Co-authored-by: Codex <noreply@openai.com>
|
2026-10-27 17:50:19 -07:00 |
|
|
Giancarlo Delfin
|
820be160ce
|
[Spec Decode] Disable batch-sharded draft logits for PCP (#64151)
Signed-off-by: Giancarlo Delfin <gdelfin@inferact.ai>
|
2026-10-27 17:00:23 -07:00 |
|
|
|
8697bbae4c
|
[Bugfix][Attention] Fix FlashInfer MLA DCP decode on ragged spec-decode batches (#59392)
Signed-off-by: zjy0516 <riverclouds.zhu@qq.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Kevin H. Luu <khluu000@gmail.com>
Co-authored-by: Pavani Majety <pmajety@nvidia.com>
|
2026-10-27 16:59:59 -07:00 |
|
|
Moein Khazraeeandvllm-agent
|
36f5ff3172
|
[Perf][KV Offload] Reduce KVCR adapter overhead (#63889)
Signed-off-by: Moein Khazraee <moein@nvidia.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
|
2026-10-27 19:41:00 -04:00 |
|
|
|
e96b490b12
|
[Bugfix][Kimi-K3] Reserve KDA prefill workspace before profiling (#63989)
Signed-off-by: jbernloehr <jbernloehr@nvidia.com>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: Thien Tran <gau.nernst@yahoo.com.sg>
Co-authored-by: vllm-agent <claw@inferact.ai>
|
2026-10-28 05:35:07 +07:00 |
|
|
Wentao Yeandvllm-agent
|
c56706e88e
|
[Perf] Use TMA for MOE up/down proj, 3.1~3.4% E2E Throughput Improvement for GLM5.3 (#63820)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
|
2026-10-27 18:21:39 -04:00 |
|
|
Jie Hao
|
0855defde8
|
Revert "[NIXL] Bump NIXL version to 1.5.1" (#63817) (#64085)
Signed-off-by: Jie Hao <jihao@nvidia.com>
|
2026-10-27 22:07:02 +00:00 |
|
|
Sherif Waly
|
4d95f222e6
|
[Feature][Core] Enable best-effort NUMA binding by default (#61258)
|
2026-10-27 22:36:02 +01:00 |
|
|
John Calderon
|
7bf13a3db1
|
[CI][Test] Adjust TRT-LLM and GDN numerical tolerances for Rubin (#64081)
Signed-off-by: John Calderon <jcalderon@nvidia.com>
|
2026-10-27 21:33:08 +00:00 |
|
|
|
70cb50dab2
|
[Feature][HiSparse] Add a platform backend for device operations (#62592)
Signed-off-by: zhenwenqi2024 <zhenwenqi_2022@qq.com>
Signed-off-by: QiuChunshuo <qiuchunshuo@huawei.com>
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: QiuChunshuo <qiuchunshuo@huawei.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
Co-authored-by: Matthew Bonanni <mbonanni@redhat.com>
|
2026-10-27 21:18:40 +00:00 |
|
|
yzong-rh
|
c4f2bb4abb
|
[Bugfix][Spec Decode] Preserve grammar-invalid drafts for rejection sampling (#61557)
Signed-off-by: Yifan Zong <yzong@redhat.com>
|
2026-10-27 23:09:32 +02:00 |
|
|
Wentao Yeandvllm-agent
|
c6f6e6a983
|
[GLM5.3 Perf] Enable dspark fp8 kv cache for draft model, 5.38 GB GPU Memory saved per rank (#63841)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
|
2026-10-27 17:08:59 -04:00 |
|
|
|
bb7ea479c6
|
[ROCm][Bugfix][Kimi-K3] Pass zero positions to the fused MLA decode q-prep kernel (#63409)
Signed-off-by: seungrokj <seungrok.jung@amd.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
|
2026-10-27 16:08:27 -05:00 |
|
|
|
274a5e19a1
|
[Perf][DSv4.1] Capture the decoder replay graphs on the model graphs' stream (#63155)
Signed-off-by: Qing <xqoasis@gmail.com>
Signed-off-by: qing.xiao <qing.xiao@bytedance.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
Co-authored-by: qing.xiao <qing.xiao@bytedance.com>
|
2026-10-27 13:45:57 -07:00 |
|
|
|
b4a0830f66
|
[Perf] Avoid redundant tensor operations during serialization (#62464)
Signed-off-by: Jakub Zakrzewski <jzakrzewski@nvidia.com>
Signed-off-by: Jakub Zakrzewski <jzakrzewski.jz@gmail.com>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: Nick Hill <nickhill123@gmail.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
|
2026-10-27 13:34:06 -07:00 |
|
|
|
815df3af10
|
[Doc] MiniMax-M3: GB200 AgentX presets (TP4/TP8) (#64114)
Signed-off-by: Ankur Singh <ankusingh@nvidia.com>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Wei Zhao <51183510+wzhao18@users.noreply.github.com>
|
2026-10-27 20:12:12 +00:00 |
|
|
|
405276b8f0
|
[Rust Frontend] Detect Lark grammars like the Python engine (#63709)
Signed-off-by: Bugen Zhao <i@bugenzhao.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
|
2026-10-27 20:07:06 +00:00 |
|
|
Moein Khazraee
|
e3883b7a7e
|
[KV Offload] Align KVCR cache recency with request prefixes (#60469)
Signed-off-by: Moein Khazraee <moein@nvidia.com>
|
2026-10-27 20:04:19 +00:00 |
|
|
|
a94c4481e0
|
[Bugfix][Rust Frontend] Bump fastokens to 0.3.5 (#63707)
Signed-off-by: Bugen Zhao <i@bugenzhao.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
|
2026-10-27 19:31:11 +00:00 |
|
|
|
e1e77a279d
|
[Bugfix] Bound prefix cache reuse by the first requested prompt score (#60960)
Signed-off-by: Ting Sun <suntcrick@gmail.com>
Signed-off-by: aoshen02 <aoshen02@users.noreply.github.com>
Signed-off-by: Nick Hill <nickhill123@gmail.com>
Co-authored-by: aoshen02 <aoshen02@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
Co-authored-by: Nick Hill <nickhill123@gmail.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
|
2026-10-27 19:26:07 +00:00 |
|
|
Roberto L. Castroandvllm-agent
|
a2ad4b74b9
|
[Perf] Enable DeepGEMM sparse MQA logits on the FP8 indexer cache, DSv4.1 (#62011)
Signed-off-by: LopezCastroRoberto <rocastro@redhat.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
|
2026-10-27 19:25:19 +00:00 |
|
|
|
2897401370
|
[Perf][KV Connector][Mooncake] Vectorize transfer descriptor build (#62478)
Signed-off-by: CJJ1008 <843515223@qq.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
|
2026-10-27 12:25:04 -07:00 |
|
|
|
9af236a332
|
[Bugfix][Perf][ROCm] RDNA launch config for kernel_paged_attention_2d (#63735)
Signed-off-by: Jipeng Li <jipengli@amd.com>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Simon Danielsson <70206058+simondanielsson@users.noreply.github.com>
|
2026-10-27 14:23:06 -05:00 |
|
|
GuorongYeandgrYe99
|
bfa2a43345
|
[CI] Fix stale NVFP4 decode test selector on DGX Spark (#64107)
Signed-off-by: grYe99 <guorongye99@gmail.com>
Co-authored-by: grYe99 <guorongye99@gmail.com>
|
2026-10-27 12:00:49 -07:00 |
|
|
|
741584f8ca
|
[Bugfix][MLA] Gather nvfp4_ds_mla chunked-context KV with the NVFP4 kernel (#58744)
Signed-off-by: Stu Cao <stu@primeintellect.ai>
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-authored-by: Lucas Wilkinson <lwilkins@redhat.com>
Co-authored-by: Lucas Wilkinson <LucasWilkinson@users.noreply.github.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
|
2026-10-27 14:35:01 -04:00 |
|
|
|
3bc521d9e6
|
[Bugfix][HiSparse][KV Connector] Fix DecodeBenchConnector crash on bounded KV cache groups (#61543)
Signed-off-by: grYe99 <guorongye99@gmail.com>
Co-authored-by: grYe99 <guorongye99@gmail.com>
Co-authored-by: Claude Haiku 5.5 <noreply@anthropic.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
|
2026-10-27 14:34:41 -04:00 |
|
|
Ankur Singh
|
a75277a0a1
|
[Doc] MiniMax-M3.1-Flash: add the B300 single-node AgentX presets (#64108)
Signed-off-by: Ankur Singh <ankusingh@nvidia.com>
|
2026-10-27 11:04:42 -07:00 |
|
|
Nicolò Lucchesi
|
4ddc7aa40f
|
[KV Connector][NIXL] Read only a group's own regions in hybrid SSM transfers (#62317)
Signed-off-by: NickLucche <nicolo.lucchesi@mistral.ai>
|
2026-10-27 19:02:22 +01:00 |
|
|
hclandvllm-agent
|
b3de57ad51
|
[CI][Bugfix] Set gpu_memory_utilization for the Spark GPQA eval config (#58232)
Signed-off-by: Chenglun Hu <chenglunhu@gmail.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
|
2026-10-27 14:01:10 -04:00 |
|
|
|
d0c20bb840
|
[ROCm][Perf][Attention] Keep the bf16 attention prefill when FP8 padding would double the heads (#61863)
Signed-off-by: Lucas Santos <Lucas.Santos@amd.com>
Co-authored-by: Rohan Potdar <rohanpotdar138@gmail.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
|
2026-10-27 17:28:37 +00:00 |
|
|
|
eaf1627c01
|
[ROCm][aiter] officially supporting aiter Triton kernels+dsv4 on RDNA3 (#52970)
Signed-off-by: xawang <xawang@amd.com>
Signed-off-by: skysnow2001 <skysnow9285@gmail.com>
Co-authored-by: skysnow2001 <skysnow9285@gmail.com>
Co-authored-by: keneoneth <keneoneth18@gmail.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
|
2026-10-27 17:01:13 +00:00 |
|
|
|
94ad4dfbae
|
[ROCm][Perf] W4A16: opt-in dense weight cache for RDNA hybrid prefill (#60710)
Signed-off-by: Zohaib Moti <zohaib.moti@amd.com>
Co-authored-by: roberteg16 <roberteg16@users.noreply.github.com>
Co-authored-by: Douglas Lehr <91553416+dllehr-amd@users.noreply.github.com>
|
2026-10-27 16:57:52 +00:00 |
|
|
|
dc4d977d97
|
[MRV2] Let the runner decide whether to profile CUDA graph memory (#63778)
Signed-off-by: Nick Hill <nickhill123@gmail.com>
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
|
2026-10-27 12:52:47 -04:00 |
|
|
|
6aa70c4bec
|
[ROCm][Perf][Kimi-K3] Fold q_b_proj and W_UK into one GEMM for the MLA decode query (#60110)
Signed-off-by: Nicholas Susanto <nicholas.susanto@amd.com>
Signed-off-by: Omar Muhammad <omar.muhammad@amd.com>
Co-authored-by: Omar Muhammad <omar.muhammad@amd.com>
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
|
2026-10-27 11:34:59 -05:00 |
|
|
|
889d635dd7
|
[Perf][ROCm][Attention] Fuse Kimi-K3 MLA chunked-context gather on AITER (#53166)
Signed-off-by: Nicholas Susanto <nicholas.susanto@amd.com>
Signed-off-by: Omar Muhammad <omar.muhammad@amd.com>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
Co-authored-by: Douglas Lehr <91553416+dllehr-amd@users.noreply.github.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
Co-authored-by: Omar Muhammad <omar.muhammad@amd.com>
|
2026-10-27 11:29:57 -05:00 |
|
|
Carl Persson
|
228b1e3d23
|
[ROCm][DSV4.1][Perf] Single-launch DSA decode candidate mask (#59668)
Signed-off-by: Carl Persson <carl.persson@amd.com>
|
2026-10-27 16:08:14 +00:00 |
|
|
Woosuk KwonandCodex
|
e3ba8d00db
|
[Bugfix][Bench] Handle packed hybrid KV and speculative recurrent state (#63873)
Signed-off-by: Woosuk Kwon <woosuk@inferact.ai>
Co-authored-by: Codex <noreply@openai.com>
|
2026-10-27 09:04:59 -07:00 |
|
|
Robert EsclapezandDouglas Lehr
|
43111fd1f0
|
[Bugfix][Spec Decode] Fix eagle draft sharing with a quantized lm_head (#53104)
Signed-off-by: Robert Esclapez Garcia <robert.garcia@amd.com>
Co-authored-by: Douglas Lehr <91553416+dllehr-amd@users.noreply.github.com>
|
2026-10-27 15:47:22 +00:00 |
|
|
|
83909ad84b
|
[ROCm][Perf] W4A16: widen skinny GEMM LDS activation loads to b128 (#57015)
Signed-off-by: Robert Esclapez Garcia <robert.garcia@amd.com>
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: Douglas Lehr <91553416+dllehr-amd@users.noreply.github.com>
|
2026-10-27 15:46:42 +00:00 |
|
|
Isotr0pyandvllm-agent
|
a9b727a612
|
[UX] Remove --hf-overrides necessity for Qwen3-ASR-ForceAligner (#63274)
Signed-off-by: Isotr0py <Isotr0py@outlook.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
|
2026-10-27 23:20:03 +08:00 |
|
|
wangxiyuan
|
ca8dbccf20
|
[Bugfix] Forward launch grid to Triton kernel overrides (#63999)
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
|
2026-10-27 14:46:07 +00:00 |
|
|
Yusen Gao
|
bf4b3b5cb8
|
[CI][CPU] Initialize DFlash test FC after loading weights (#63973)
Signed-off-by: Yusen Gao <amenomorikou@gmail.com>
|
2026-10-27 22:41:31 +08:00 |
|
|
Nicolò Lucchesi
|
939ff73503
|
[Perf] Drop BasevLLMParameter's Python __torch_function__ (#63785)
Signed-off-by: NickLucche <nicolo.lucchesi@mistral.ai>
|
2026-10-27 15:30:25 +01:00 |
|
|
|
803bc13cf9
|
[Rust Frontend] Add --reasoning-effort-map to rewrite request reasoning efforts (#62312)
Signed-off-by: Robert Shaw <robertgshaw2@gmail.com>
Signed-off-by: Bugen Zhao <i@bugenzhao.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Bugen Zhao <i@bugenzhao.com>
|
2026-10-27 22:04:08 +08:00 |
|
|
|
4216a234cf
|
[Rust Frontend] Add per-request speculative decoding metrics (#61877)
Signed-off-by: Bugen Zhao <i@bugenzhao.com>
Co-authored-by: Bugen Zhao <i@bugenzhao.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: vllm-agent <claw@inferact.ai>
|
2026-10-27 22:03:10 +08:00 |
|