Commits vllm-project/vllm

p

khluu committed 70 days ago

f8430dc9

fix(glm47): improve tool call parsing and content normalization (#37386)

karanb192 committed 70 days ago

Verified fad09e8a

[LoRA] Make LoRA respect `language_model_only` (#37375)

jeejeelee committed 70 days ago

Verified 8c31f47c

[Bugfix] Avoid OpenMP thread reallocation in CPU torch compile (#37391)

bigPYJ1151 committed 70 days ago

Verified 26180124

[kv_offload+HMA][0/N]: Support block-level preemption handling (#34805)

orozery committed 70 days ago

Verified fcf0687b

[XPU] skip unsupported ut and update test_nixl_connector (#37179)

zhenwei-intel committed 70 days ago

Verified 86b7e3c9

[responsesAPI] parser.extract_response_outputs can take in token IDs (#37130)

qandrew committed 70 days ago

Verified 0e959161

[CI] Stabilize test_cpu_offloading by waiting for async offload before cache reset (#37335)

AndreasKaratzas committed 70 days ago

Verified ce2ef42f

[ROCm][CI] Add ROCM_EXTRA_ARGS to audio_in_video test server fixture (#37349)

AndreasKaratzas committed 70 days ago

Verified 8b632575

[Hardware][TPU] Add supports_async_scheduling() method to Executor interface so that it can be extended for Executor implementations. (#36924)

gxd3 committed 70 days ago

Verified a0dd1995

[Perf] Enable dual stream execution of input projection for Qwen3 (#36795)

xyang16 committed 70 days ago

Verified f1740006

[ROCm][CI] Skip trtllm kvfp8 dequant tests on ROCm (#37330)

AndreasKaratzas committed 70 days ago

Verified 58cde5c0

[Performance] Add --enable-ep-weight-filter CLI option (#37351)

esmeetu committed 70 days ago

Verified 761e0aa7

[Kernel][Helion] [16/N] Refactor register_kernel API to be more Dynamo-friendly (#36705)

gmagogsfm committed 70 days ago

Verified ff9fbc9a

[ROCm][Quantization] add fp8xfp8 attn support for rocm_aiter_unified_attn (#36927)

divakar-amd committed 70 days ago

Verified e6c47977

[Kernel] Add non-gated support for NVFP4 CUTLASS MoE (#37320)

mgoin committed 70 days ago

Verified 09e4576f

[ROCm] Validate block_size for explicitly selected attention backends (#36846)

AndreasKaratzas committed 70 days ago

Verified 3ed7b1e6

[Bugfix][ROCm] Fix worker startup OOM on ROCm by skipping unreliable cudagraph memory profiling (#36720)

JartX committed 70 days ago

Verified e8f9dbc3

Make KV connector metadata build overridable via plugin (#37336)

sarckk committed 71 days ago

Verified de35c06c

[Model] Add ColQwen3.5 4.5B support (#36887)

athrael-soju committed 71 days ago

Verified c0745a85

[Models] Cohere ASR (#35809)

ekagra-ranjan committed 71 days ago

Verified b5ca9c35

[Bugfix] Rescale NVFP4 weight scales to fix BF16 dequant underflow (#34577)

ricky-chaoju committed 71 days ago

Verified 24575899

[Bugfix] Fix mock.patch resolution failure for standalone_compile.FakeTensorMode on Python <= 3.10 (#37158)

dbari committed 71 days ago

Verified 1204cf0a

[Perf] Set Flashinfer sparse MLA as default backend for FP8 kv cache (#37252)

wzhao18 committed 71 days ago

Verified b36adfa3

[Deprecation] Deprecate `--calculate-kv-scales` option (#37201)

mgoin committed 71 days ago

Verified e78821b4

[Model] Remove unused `handle_oov_mm_token` (#37321)

DarkLight1337 committed 71 days ago

Verified 51f0acda

bump compressed-tensors version to 0.14.0.1 (#36988)

brian-dellabetta committed 71 days ago

Verified fa75204b

[Bug] Fix FlashInfer MNNVL socket collisions under concurrent vLLM jobs (#36674)

yewentao256 committed 71 days ago

Verified bdb903bb

[Torch 2.11] Guard torch._C._cpu attribute checks for forward compatibility (#35673)

atalman committed 71 days ago

Verified 68f783a7

[CI] Split Distributed Tests (4 GPUs) and Kernel MoE tests (#37100)

avinashsingh77 committed 71 days ago

Verified c5030c43