vllm-project/vllm

Pull Requests Commits

[Quantization] Bound peak memory when repacking FP4 MoE weights for Marlin (#47851)

joerowell committed 53 days ago

Verified 1bf3997e

Add VLLM_FLASHINFER_AUTOTUNE_SKIP_OPS and skip CuTeDSL fp4_gemm autotuning by default (#48268)

mgoin committed 53 days ago

Verified 29fd6888

[Bugfix] Fix thinking_token_budget not enforced after natural </think> re-entry (#45984)

ashwing committed 53 days ago

Verified ed908cf0

[Bugfix][Test] Register Qwen/Qwen3.5-4B example model (#48276)

njhill committed 53 days ago

Verified 26ff616b

handle topk_ids padding in align sum kernel (#47785)

gnovack committed 53 days ago

Verified f378f79b

[Bugfix] Fix FlashMLA dense fp8 metadata crash (num_sm_parts clamp) (#48045)

MatthewBonanni committed 54 days ago

Verified 735def4f

[ROCm] Enable DeepSeek-V4 DSpark speculative decoding on AMD (MI350X / MI355X, gfx950) (#47419)

larryli2-amd committed 54 days ago

Verified c227aaa3

[Model] Add LongCat-Flash-Lite (n-gram embedding) (#47857)

mgoin committed 54 days ago

Verified 08dfd686

[Build/CI] Build arm64 PR and postmerge image builds for Blackwell SM10x and SM110 (#48041)

tlrmchlsmth committed 54 days ago

Verified 978a6dfa

fix: correct load_weights track logic and enable weight integrity for… (#41811)

MynameFelix committed 54 days ago

Verified 85c09e98

[Bugfix] Fix turboquant FP8 cast failure for BF16 models on Ampere GPUs (#39988)

XuZhou26 committed 54 days ago

Verified b12cca6a

[Refactor] Remove unused rocm kernel `combine_topk_swa_indices_ragged` (#48158)

yewentao256 committed 54 days ago

Verified e257faf8

[Model] Migrate MistralLarge3ForCausalLM to AutoWeightsLoader (#48153)

Functionhx committed 54 days ago

Verified fabec87f

[Bugfix][Spec Decode] Fix DFlash draft/target layer-count mismatch (#48113)

njhill committed 54 days ago

Verified 7614b88e

[Misc] Remove dead code in ViT functionality test (#48220)

Isotr0py committed 54 days ago

Verified 68ea76e7

[Rust Frontend] Add roundtrip fixtures for more chat parsers (#47883)

reidliu41 committed 54 days ago

Verified c241c7a2

Deepstream video backend (#42424)

ViranjanPagar committed 54 days ago

Verified e23b1930

[CI] Add TORCH_NIGHTLY=1 build mode (run full suite on torch nightly) (#47180)

atalman committed 54 days ago

Verified f36284a8

[Model][CI/Build] Cosmos3: enable registry tests and register Cosmos3-Super (#48211)

guoriyue committed 54 days ago

Verified 424df4f6

[Rust Frontend] Integrate MM video support (#47959)

BugenZhao committed 54 days ago

Verified 074bdd0d

Add XPU nightly and release image publishing to DockerHub (#48126)

wendyliu235 committed 54 days ago

Verified 216ee587

[CI] Right-size test-area timeouts from nightly durations (#48186)

khluu committed 54 days ago

Verified 433f2911

[XPU] Enable v1/sample tests on XPU CI (#44472)

chaojun-zhang committed 54 days ago

Verified 28eaf05d

[Perf] fuse more rmsnorm and all-reduce in qwen3.5 (#46998)

ZJY0516 committed 54 days ago

Verified 300e3379

[Feature][Parser] Support include_reasoning param for non-Harmony models (#44301)

albertoperdomo2 committed 54 days ago

Verified 5715fde1

[Core][KV events] Report prefix-cache-reused blocks in full report mode (#45261)

GongLei-HW committed 54 days ago

Verified e5588e49

DCP supports hybrid attention (#40996)

Yancey0623 committed 54 days ago

Verified 95ed0fea

[kv_offload] Emit tier-owned BlockStored events from FS/OBJ secondary tiers (#47923)

Change72 committed 54 days ago

Verified 2d814a00

[CI/Build][AMD] Fix ROCm OOM in eagle_correctness_heavy by reserving CUDA graph memory (#47366)

peizhang56 committed 54 days ago

Verified 88e5e2c5

[bugfix] bge-m3-sparse-plugin mismatch requests (#48112)

staugust committed 54 days ago

Verified feb384ad

Newer Older