vllm-project/vllm

Pull Requests Commits

EPLB: add post-rearrangement weight invariant checks

tlrmchlsmth committed 37 days ago

36486dd0

[BugFix] Fix EPLB balancedness metric using wrong dimension

tlrmchlsmth committed 37 days ago

1030f4c8

Fix EPLB + NVFP4 modelopt: use direct Parameter assignment for g1/g2_alphas

tlrmchlsmth committed 37 days ago

36b60baf

Fix EPLB + NVFP4: exclude broadcast scales and fix stale quant config

tlrmchlsmth committed 37 days ago

4f57aa65

Adding support to Sarvam's MoE models (#33942)

rahul-sarvam committed 37 days ago

Verified 85f50eb4

[Misc] Remove duplicate parser registration (#36303)

taneem-ibrahim committed 37 days ago

Verified 5261223c

[V0 Deprecation] Remove unused swap_space parameter (#36216)

majiayu000 committed 37 days ago

Verified 00b814ba

[Bugfix] Fix compressed-tensors quantization failure for DeepSeek-R1 on MI300x (#36247)

vllmellm committed 37 days ago

Verified ee8a2951

feat: expose media_io_kwargs at runtime (#34778)

milesial committed 38 days ago

Verified 755356b3

[ROCm][CI] Making entrypoints more deterministic on ROCm (#36293)

AndreasKaratzas committed 38 days ago

Verified 58928475

Fix CUDA graph decode capture crash in AITER FlashAttention (#36042)

iseeyuan committed 38 days ago

Verified 1a971808

refine `vllm bench throughput --backend hf` (#35971)

jikunshang committed 38 days ago

Verified 7eb524e6

[BugFix] Avoid ignored trust_remote_code warnings (#36290)

njhill committed 38 days ago

Verified c7f32e08

[Model Runner V2] Fix warmup for pipeline parallel (#36280)

njhill committed 38 days ago

Verified b3546865

[Core] Fix benign error log during normal shutdown (#36270)

njhill committed 38 days ago

Verified 6a18d878

mla: don't update kv cache on dummy forwards (#36282)

itayalroy committed 38 days ago

Verified 24a03915

[ROCm][CI] Fixing yaml file for external amd-ci signal (#36284)

AndreasKaratzas committed 38 days ago

Verified b5e34e1f

[docs][torch.compile] Add fusions.md — kernel/operator fusion reference page (#35538)

Copilot committed 38 days ago

Verified ce8546a1

[ROCm] Support MLA with nhead<16 and FP8 KV cache for TP=8 (Kimi K2.5/Linear) (#35850)

ChuanLi1101 committed 38 days ago

Verified c188749b

Enabling some B200-specific tests on MI355 (#35253)

Alexei-V-Ivanov-AMD committed 38 days ago

Verified 225d1090

[CustomOp] CustomOp FusedRMSNormGated (#35877)

eellison committed 38 days ago

Verified f3c6c9c9

Revert "[BugFix] Fix engine hanging after KV cache initialization fai… (#36262)

njhill committed 38 days ago

Verified 26bd43b5

[Bugfix] Quickfix followups to busy loop removal in #28053 (#36068)

tjohnson31415 committed 38 days ago

Verified 6b625a88

[compile] Stop unconditionally patching constrain_to_fx_strides (#36152)

zou3519 committed 38 days ago

Verified 54756b61

[Bugfix] Fix `cudagraph_mode:FULL` dispatch (This does not impact `FULL_AND_PIECEWISE` (default)) (#36165)

TQCB committed 38 days ago

Verified 39f9ea0d

[Refactor] Modular video loader backend refactoring (#35202)

Isotr0py committed 38 days ago

Verified e4ae148a

[Misc] Lazy import registered processors (#36024)

Isotr0py committed 38 days ago

Verified 1d0c0d20

[bugfix] add api process rank in default multimodal request (#36150)

fake0fan committed 38 days ago

Verified fcb73f30

[CI] Fix startup error test (#36230)

hmellor committed 38 days ago

Verified e2090bf3

[CI][MM] Gate vision encoder attention mask to MiniCPM only, fixing Aria regression (#36206)

AndreasKaratzas committed 38 days ago

Verified 2a00d324

Older