[Hardware][TPU][V1] Multi-LoRA implementation for the V1 TPU backend (#14238)

Akshat Tripathi committed over 1 year ago

[Quantization] Quark MXFP4 format loading (#16943)

Bowen Bao committed over 1 year ago

[Bugfix] Fix missing lora name mapping for lora without prefix (#17793)

Isotr0py committed over 1 year ago

Fix and simplify `deprecated=True` CLI `kwarg` (#17781)

Harry Mellor committed over 1 year ago

[Bugfix] Fix Video IO error for short video (#17791)

Isotr0py committed over 1 year ago

Only depend on importlib-metadata for Python < 3.10 (#17776)

Christian Heimes committed over 1 year ago

[ROCm][FP8][Kernel] FP8 quantization fused into Custom Paged Attention (#17139)

Gregory Shtrasberg committed over 1 year ago

[doc] update the issue link (#17782)

Reid committed over 1 year ago

Make key optional for rotary embedding (#17566)

Yong Hoon Shin committed over 1 year ago

Fix test_memory_usage_no_spec (#17754)

Yong Hoon Shin committed over 1 year ago

[Frontend] Add missing chat templates for various MLLMs (#17758)

Cyrus Leung committed over 1 year ago

Add NeuronxDistributedInference support, Speculative Decoding, Dynamic on-device sampling (#16357)

Satyajith Chilappagari committed over 1 year ago

[Misc] Remove qlora_adapter_name_or_path (#17699)

Jee Jee Li committed over 1 year ago

[Kernel] Use fused rmsnorm for some models like qwen3 series (#17735)

Wanrui Dai committed over 1 year ago

[Kernel] GGUF MoeVec kernel (#16780)

Szymon Ożóg committed over 1 year ago

[Misc] Use `apply_rotary_emb` from vllm_flash_attn for Qwen2-VL vision RoPE (#17726)

Isotr0py committed over 1 year ago

[Misc] Split model loader (#17712)

Jee Jee Li committed over 1 year ago

[BugFix][Spec Decode] Fix hidden size mismatch between target and eagle head (#17740)

Woosuk Kwon committed over 1 year ago

Replace lm-eval bash script with pytest and use enforce_eager for faster CI (#17717)

Michael Goin committed over 1 year ago

[TPU] Add kernel test for moe_pallas (#17496)

Michael Goin committed over 1 year ago