Use Transformers v5 `WeightRenaming` for Transformers modeling backend (#31545)

Harry Mellor committed 8 months ago

[Bugfix][Spec Decode] Avoid double call of Ngram CPU (#36952)

Ekagra Ranjan committed 8 months ago

[ROCm][Quantization] add quark w4a8 mxfp4_fp8 for LinearLayer (#35316)

Divakar Verma committed 8 months ago

[Frontend][Core] Re-add shutdown timeout - allowing in-flight requests to finish (#36666)

Mark McLoughlin committed 8 months ago

[Misc] Set default `kv_buffer_device` in a better way (#36862)

Harry Mellor committed 8 months ago

Fp8 lora dense kernel (#35242)

yugong333 committed 8 months ago

[Bugfix] Fix Qwen2.5-omni/Qwen3-omni mm_processor cache for audio_in_video request (#36800)

Isotr0py committed 8 months ago

[Bugfix] fix paddleocr crash on some image shape (#36959)

bigmoyan committed 8 months ago

[2/N] Elastic EP Milestone 2: Integrating NIXL-EP (#35627)

Itay Alroy committed 8 months ago

[XPU] Support LoRA via torch.compile on XPU platform (#36962)

Chaojun Zhang committed 8 months ago

[ROCm][CI] Upgrading orchestrator to handle python pipeline markers and options (#36181)

Andreas Karatzas committed 8 months ago

[kv_offload+HMA][1/N]: Support multiple KV groups in OffloadingSpec (#36610)

Or Ozeri committed 8 months ago

[ROCm][CI] Corrected the GPT-OSS test root path (#36711)

Andreas Karatzas committed 8 months ago

[Frontend] Delegate preprocessing to `OpenAIServingRender` (#36483)

Sage committed 8 months ago

Enable RoPE+KV cache fusion for ROCm AITER FA (non-shuffle layout) (#35786)

Rohan Potdar committed 8 months ago

[Tests] Shutdown test `RemoteVLLMServer` cleanly (#36950)

Nick Hill committed 8 months ago

[Bugfix] Fix FlashInfer GDN warmup ValueError on SM90 GPUs (#36876)

Thomas Parnell committed 8 months ago

[Frontend] Fix usage incorrectly returned with empty stream_options` (#36379)

Csrayz committed 8 months ago

fix(kv-cache): increase hybrid attention grouping threshold from 1.25 to 1.5 (#36684)

jaime campos salas committed 8 months ago

[Model] Add support for BERT-like Chinese ERNIE pooling models (#36385)

whyiug committed 8 months ago