vllm-project/vllm

Pull Requests Commits

[KV offload][5/N] Add `CPUOffloadingSpec` (#24251)

orozery committed 104 days ago

Verified 8db29392

[torch.compile] Cleanup compilation tests and custom passes, add debug utils, fix DCE bug (#23091), fix test (#24376), and prep for custom op matching (#24604) (#24542)

ProExpertProg committed 104 days ago

Verified d5e0fca2

[misc] Remove RFC review hours reference (#25416)

simon-mo committed 104 days ago

Verified 8d0ee5a5

[DP] support torchrun external launcher with Data Parallelism (#24899)

luccafong committed 104 days ago

Verified 922979bf

[CI Failure] Fix fp8 kv cache on <SM90 (#25396)

mgoin committed 104 days ago

Verified 239ef0c1

[Compiler] Disable Inductor standalone compile by default (#25391)

ElizaWszola committed 104 days ago

Verified 1d7f95b8

[CLI env var] Add VLLM_FLASH_ATTN_MAX_NUM_SPLITS_FOR_CUDA_GRAPH in env variables (#25274)

Daisy-Ma-coder committed 104 days ago

Verified cfbee3d0

[EPLB] Reduce EPLB Inference Overhead (#24573)

abmfy committed 104 days ago

Verified 06a41334

[V1][Attention] Split triton_attn in triton-only and rocm specific backends (#24648)

bringlein committed 104 days ago

Verified 175811e3

[Bugfix] Fix several issues with p2p xPyD in GET type (#23993)

Csrayz committed 104 days ago

Verified c10101a3

[Kernel] MI-300X triton moe configs (#23445)

Sara-KS committed 104 days ago

Verified ac243886

Make `mypy` behave like a proper pre-commit hook (#25313)

hmellor committed 105 days ago

Verified 3d2c56b7

Make pickle import check fast (#25379)

hmellor committed 105 days ago

Verified 64c824cd

[Misc] Remove unused encoder-decoder error strings (#25374)

DarkLight1337 committed 105 days ago

Verified 417a164a

refactor: abstract graph mode support into platform interface (#25161)

yiz-liu committed 105 days ago

Verified b6f01bd9

[TPU] Deprecate `xm.mark_step` in favor of ``torch_xla.sync` (#25254)

NickLucche committed 105 days ago

Verified 4cf71cc8

[TPU][Bugfix][CI] Fix broken tests/build dependency (#25255)

NickLucche committed 105 days ago

Verified a66d1313

Enable Eagle3 speculative decoding for GPT-OSS model (#25246)

eldarkurtic committed 105 days ago

Verified 21467f9a

[V0 Deprecation] Remove `MultiModalPlaceholderMap` (#25366)

DarkLight1337 committed 105 days ago

Verified f92d9526

[V0 Deprecation] Remove V0-only methods in multi-modal registry (#25362)

DarkLight1337 committed 105 days ago

Verified 6d0b827c

[Bugfix] Fix hermes tool parser handling of non-string argument types (#22002)

david6666666 committed 105 days ago

Verified 0eecb316

[Docs] GSM8K Accuracy Evaluation doc update (#25360)

david6666666 committed 105 days ago

Verified 793be8d0

[Model] Support Dots OCR (#24645)

ywang96 committed 105 days ago

Verified 7b57a433

Multimodal - audio tests (#25285)

debroy-rh committed 105 days ago

Verified 5aeb9254

[Bugfix][V0 Deprecation][CI] use async mock and await for async method (#25325)

KKSK-DON committed 105 days ago

Verified 04d37523

Remove V0 attention backends (#25351)

WoosukKwon committed 105 days ago

Verified bc6e542d

[Perf] Further optimization for Qwen3-VL `fast_pos_embed_interpolate` (#25347)

Isotr0py committed 105 days ago

Verified af7dfb0d

[V0 Deprecation] Remove V0 sampling metadata (#25345)

WoosukKwon committed 105 days ago

Verified 1c3ffdbe

feat: Enable engine-level arguments with speculators models (#25250)

rahul-tuli committed 105 days ago

Verified c438b295

[V0 Deprecation] Remove async_output_proc, preemption mode, delay factor (#25334)

WoosukKwon committed 105 days ago

Verified 0ff8ebb2

Older