Commits vllm-project/vllm

[Bench] Pad vllm bench forward batches as serving's CUDA graphs would

WoosukKwon committed 23 hours ago

7e61bcca

[Bench] Time target sampling in vllm bench forward (#66483)

WoosukKwon committed 23 hours ago

Verified 871473fa

[Bugfix] Map compressed-tensors KV-cache scale names for GPT-J (#65949)

HollowMan6 committed 23 hours ago

Verified 3ac8ba9f

[Rust Frontend] Keep special tokens' spelling in logprob tokens (#66477)

BugenZhao committed 23 hours ago

Verified 3a092abe

[Bugfix][Multimodal] Fill the Qwen2-VL video budget up to the context length (#66432)

subhashpolisetti committed 1 day ago

Verified 821851ea

[ROCm][Spec Decode] Enable fused multi-step draft decode for ROCM_AITER_UNIFIED_ATTN (#63096)

fengshaoyi-amd committed 1 day ago

Verified fdd90995

[Frontend] Mark disabled scale-out routes in the startup route log (#66481)

aoshen02 committed 1 day ago

Verified 343b5a56

[Bugfix][LoRA] Use 3D MoE LoRA wrapper for Gemma4 (#64437)

gc-fu committed 1 day ago

Verified b6ed63d2

[CI][KV Connector] Shard the NIXL PD accuracy core sweep and use decode-only CUDA graphs (#66412)

ZJY0516 committed 1 day ago

Verified 116025c6

[XPU] Bump up kernel 0.1.17, change MoE and MLA interfaces (#60609)

ZhouXiaoya25 committed 1 day ago

Verified b0fd6874

[Spec Decode][PP] Centralize auxiliary hidden-state relay (#66177)

xsxszab committed 1 day ago

Verified b936e142

[CPU][recipes] Add Qwen3-VL-4B-Instruct support for Intel Xeon 6 (#66238)

louie-tsai committed 1 day ago

Verified c7cda53c

[Bugfix][Spec Decode] Propagate rope hf_overrides to MTP draft (#58094)

gokay-ai committed 1 day ago

Verified 083fbac0

[Rust Frontend] support --enable-force-include-usage (#60274)

lianakoleva committed 1 day ago

Verified 518dc3c9

[Bugfix][Model] Fix Inkling NVIDIA sconv cache block-size mismatch (#51951)

fjosw committed 1 day ago

Verified 21ef2e5b

[XPU][CI]Migrate several B70 jobs from intel_gpu to intel_gpu_k8s (#66034)

zxd1997066 committed 1 day ago

Verified bf19789e

[Bugfix][Qwen4Exp] Exclude the HC padding shard from the loaded-shard check (#66461)

gau-nernst committed 1 day ago

Verified 000a53c2

[CI][ROCm] Keep FLA's shared memory probe from initializing HIP (#66434)

AndreasKaratzas committed 1 day ago

Verified 3f6f0fd0

[Core][Kimi K3.1] KV cache and attention changes to serve Kimi K3.1 at TP8 (Kimi K3.1 split 3/5) (#66037)

ZJY0516 committed 1 day ago

Verified 4088c283

[Bugfix][CPU] Name the cause when a DP rank's assigned NUMA nodes don't exist (#64499)

amd-lalithnc committed 1 day ago

Verified feed3a44

[MoE] Make fused MoE routers pluggable (#51006)

freyfwt committed 1 day ago

Verified 199c2e4b

[CI] Fix docstring style in benchmark_batched_moe.py (#66441)

AndreasKaratzas committed 1 day ago

Verified 3213796c

[Bugfix][ROCm] Let the FP8 MoE oracle fall back when the global AITER flag can't serve the layer (#66350)

AndreasKaratzas committed 1 day ago

Verified 62902f6c

[Sleep Mode] Remove level-2 workarounds that #66010 makes unnecessary (#66287)

aoshen02 committed 1 day ago

Verified e5f2ea9b

[ROCm][Test] Cover the MXFP4 W4A16 intermediate pad that reaches AITER (#66206)

AndreasKaratzas committed 1 day ago

Verified 3b33a845

[XPU] All-to-all Expert-Parallellism backend for batched MoE (#46871)

oonyshch committed 1 day ago

Verified 05553901

[Bugfix][ROCm] Size the extensible KV cache from exact VRAM use, not 2 MB-rounded free memory (#63401)

GodRishUniverse committed 1 day ago

Verified fd0c471b

[Deps] Bump tensorizer to 2.12.1 for Python 3.14 (#66313)

AndreasKaratzas committed 1 day ago

Verified 5d840812

[DeepSeek-V4] Support sequence parallelism with pipeline parallelism for MegaMoE (#65766)

xsxszab committed 1 day ago

Verified 15166efa

[CI] Add check-docstring-style pre-commit hook and use single backticks in docstrings (#65158)

hmellor committed 1 day ago

Verified 9b9bd049