vllm-project/vllm

Pull Requests Commits

Merge branch 'main' into wentao-remove-dead-code-multiplaces

yewentao256 committed 9 hours ago

Verified b3bd99a9

[Perf][glm5next] Add SM121 TP=2 skinny-GEMM plans for GLM-5.3-Flash (#62956)

TimRen01 committed 9 hours ago

Verified e5196e29

[Spec Decode][ROCm] draft_lm_head_quantization for DSpark drafters (#64368)

AMD-yanfeiwang committed 13 hours ago

Verified 7b77ed58

[Misc] Allow weights mapper to transform weights (#64869)

Isotr0py committed 14 hours ago

Verified 87cb835d

[Bugfix][Model] Make compile_mm_encoder work for RADIO/NaViT encoders (#63415)

HangruiCao2000 committed 14 hours ago

Verified 10da92ff

[MORI] Add MORI EP V2 support (#61361)

ChengYM5685 committed 14 hours ago

Verified d5301a4f

[Perf][Qwen3.8] Add SM103 skinny GEMM plans for TP=1 (#64650)

namgyu-yun committed 16 hours ago

Verified 900b8b08

[RL][Model Loader] Direct weight reload: write checkpoint tensors into the live model (#64336)

hao-aaron committed 17 hours ago

Verified b5755c3b

[Model] Add NVIDIA Canary (nvidia/canary-1b-v2) speech-to-text support (#64215)

Alex-Wengg committed 17 hours ago

Verified 886b845e

[KV Connector][NIXL] Pair KV regions by layer for every transfer (#63416)

Dao007forever committed 18 hours ago

Verified 8b55f919

[Bugfix][gRPC] Drain in-flight RPCs and engine work within --shutdown-timeout (#62967)

mrn3088 committed 18 hours ago

Verified 5a233286

[Feature][DSv4.1] Enable PCP with PP or DP (#63098)

wangyicong52 committed 19 hours ago

Verified 2b009fc6

[Perf][GLM-5.3-Flash] Update FlashKDA pin, consume row-strided QKV directly (#61614)

qianlihuang committed 20 hours ago

Verified 60281154

[NIXL] Bump NIXL version to 1.5.2 (#64579)

erezzarum committed 21 hours ago

Verified 06f36c12

[Bugfix][ROCm] Size AITER MLA persistent metadata for every decode query length (#63116)

BaoYunkai committed 21 hours ago

Verified c4f54cf5

[ROCm][Kimi-K3] Keep KDA prefill checkpoints with the gluon backend (#63297)

seungrokj committed 21 hours ago

Verified e3eec515

[ROCm][DSv4.1][Perf] Enable prefix-cache SWA replay on gfx950 (#64094)

karverma-amd committed 21 hours ago

Verified 138dfa54

[Kernel] Qwen4Exp: compute HyperConnection and PLE gate row offsets in int64 (#64820)

seanbell committed 21 hours ago

Verified 58f245c6

[Pooling] Speed up ColBERT reranking with batched token heads (#64269)

taneem-ibrahim committed 21 hours ago

Verified 16332714

[Perf][LoRA][MoE] Use the shrink/expand kernels for MoE-LoRA batches of up to 64 tokens (#64653)

aoshen02 committed 21 hours ago

Verified 042f30f6

[Bugfix][Parser] Parse closed Qwen3 tool parameters once while streaming (#64675)

vMaroon committed 22 hours ago

Verified c7a683bd

[Perf][Parser] Skip tool-argument conversions after a replaced value (#64673)

vMaroon committed 22 hours ago

Verified 0aa72a45

[Frontend] Reject Anthropic-defined tools by type in /v1/messages (#64575)

robertgshaw2-redhat committed 22 hours ago

Verified 5829ca82

[Perf][Frontend] Decode streamed custom-tool input incrementally (#64674)

vMaroon committed 22 hours ago

Verified 8777b4c8

[Feature] Default EPLB num_redundant_experts to minimum valid value (#30614)

majiayu000 committed 22 hours ago

Verified 8ecf86ba

[Config] Move V2-only checks into V1 model runner unsupported features (#64830)

njhill committed 22 hours ago

Verified 1944a235

[ROCm][Model] Step-5-Preview support on ROCm (#61837)

akii96 committed 23 hours ago

Verified 17550580

[Doc] MiniMax-M3.1-Flash: add GB300 InferenceX AgentX presets (#64821)

hshrivastava-droid committed 23 hours ago

Verified d90e35cf

[ROCm][CI] Run MI355 MoE Kernels on a full MI355 instead of DPX (#64740)

djramic committed 1 day ago

Verified 59d58fd2

[Bugfix] Fix multi-layer MTP module selection for Qwen3.5 and Qwen3-Next (#62111)

spandantiwari committed 1 day ago

Verified dd5f5eeb

Older