Commits vllm-project/vllm

[Kernel] Add B12X dense linear backends (#52016)

lukealonso committed 78 days ago

Verified 3c79b1a8

[XPU] [Bugfix] process ragged weights in xpu linear backend (#52118)

zufangzhu committed 78 days ago

Verified fe4c5dcd

[Bugfix] Reject negative token ids as out-of-vocabulary (#51795)

ECMGit committed 78 days ago

Verified b216db3e

[Kernel] Gemma-4 FA4 FP8 Kernel (#48666)

jhaotingc committed 78 days ago

Verified 827a2af8

[Bugfix] Reject NUL byte in structured_outputs.regex (#51796)

ECMGit committed 78 days ago

Verified 38f097fa

[Model Runner V2][Spec Decode] Add KV cache support for multi-layer MTP (#50062)

TheEpicDolphin committed 78 days ago

Verified f80b66f5

[Bugfix] Fix .../mrope.py::apply_interleaved_rope() when torch.compile is used in torch==2.13 (#52005)

bastefaniak committed 78 days ago

Verified 71b0da7c

[Bugfix] Reapply 50869 (#52223)

benchislett committed 78 days ago

Verified 51def784

[Bugfix] Correct prompt lengths for timed_traces benchmark (#45423)

s3woz committed 78 days ago

Verified 63550515

[Attention] Fix FlashInfer SM12x prefill with sinks (#52148)

askliar committed 78 days ago

Verified b652dedd

[CI Failure] Fix CUDA wheel build for the Kimi K3 fused MLA kernel (#52210)

mgoin committed 78 days ago

Verified 2e2ffd10

[Platform] Add check_runner_kv_caches_multi_layer (#51633)

wangxiyuan committed 78 days ago

Verified 73b83949

[Bugfix][ROCm][CI] Give the AITER MLA decode metadata stub its MLA dims (#52139)

stefankoncarevic committed 78 days ago

Verified 11c3fa4a

[ROCm] Defer `tilelang` import through its import `from vllm.tilelang_utils import tilelang` and relaxed `has_tilelang` (#51159)

fxmarty-amd committed 78 days ago

Verified c5b7c069

[Bugfix][Structured Output] Mask request stop tokens in xgrammar until grammar terminates (#50595)

yzong-rh committed 78 days ago

Verified e6b2a8ad

[Quantization] Remove dead `QuantizationConfig.is_mxfp4_quant` (#51793)

fxmarty-amd committed 78 days ago

Verified 48825acc

[ROCm][CI] Solidify entrypoint LLM lifecycle (#51280)

AndreasKaratzas committed 78 days ago

Verified b96bcd0b

Auto-ping Cohere on related issues (#52091)

DarkLight1337 committed 79 days ago

Verified 69d4c3a0

[Bugfix] Declare SupportsEagle3 on KimiLinearForCausalLM (#52171)

nickus committed 79 days ago

Verified 83d4c619

Apply logit softcapping in Transformers modelling backend (#52173)

hmellor committed 79 days ago

Verified b245d8e7

[Kimi-K3] Add GEMM-RS for sequence parallelism (#52079)

gau-nernst committed 79 days ago

Verified 6014f9e6

[Bugfix][MiniCPM-V] Fix AssertionError in get_dummy_mm_data when passing VideoDummyOptions to _get_dummy_images (#52122)

mayuyuace committed 79 days ago

Verified 96acd473

[Bugfix] Fix `--data-parallel-start-rank 0` being treated as unset in `create_engine_config` (#47692)

syedalijaseem committed 79 days ago

Verified 64ca614f

Standardise weight tying on `ParallelLMHead.tie_weights` (#52147)

hmellor committed 79 days ago

Verified 80d6d557

[Frontend] Add routed-experts prompt offset (#51906)

aoshen02 committed 79 days ago

Verified 7553aac7

[ROCm] Enable V2 model runner for Kimi-K3 on ROCm (#51653)

vllmellm committed 79 days ago

Verified f3c16389

[Bugfix] Fix packed GDN decode launch for large batch-head grids (#52030)

mgoin committed 79 days ago

Verified 2d24355e

[Bugfix] Disable sequence parallelism for Dots3 NOTE (#52172)

KurodaKanbei committed 79 days ago

Verified 170592a9

[LoRA][Gemma4] Support vision tower LoRA (#42662)

linitra24 committed 79 days ago

Verified 95c91444

[Model] [Quantization] Add Ling hybrid MXFP4 routed experts support (#52114)

zexplorerhj committed 79 days ago

Verified 8e1131e1