[Bugfix] Fix Qwen3.5 config loading (#34554)

Roger Wang committed 9 months ago

[Misc] Update tests and examples for Prithvi/Terratorch models (#34416)

Christian Pinto committed 9 months ago

[new model] add COLQwen3 code & Inference (#34398)

Kata Coder committed 9 months ago

[CI] Heavy refactoring of Voxtral multimodal audio model tests (#34294)

Andreas Karatzas committed 9 months ago

Add explicit validation error for tool calls. (#34438)

Julien Denize committed 9 months ago

fix: use `__annotations__` instead of `get_type_hints()` for dynamic `kwargs` detection (#34527)

Christian S. Perone committed 9 months ago

[bug] Make sure get_modality_with_max_tokens is deterministic (#34533)

Shiyan Deng committed 9 months ago

[Feature][Perf] Support Selective CPU Weight Offloading (#34535)

Wei Zhao committed 9 months ago

[Bugfix] Fix ROCm UVA CPU weight offloading broken by #32993 (#34543)

Andreas Karatzas committed 9 months ago

[Hybrid] Enable spec decoding in mamba cache align mode (#33705)

Harry Huang committed 9 months ago

[Bugfix]: Fix structured output in multi-turn gpt-oss (#34454)

Ben Browning committed 9 months ago

Revert "[Bugfix] Fix fused MoE IMA (sans chunking) by using int64 for strides" (#34530)

Michael Goin committed 9 months ago

[Misc] vLLM's --enforce-eager should turn off compile and cudagraphs only (#34523)

Richard Zou committed 9 months ago

[Bugfix] Replace c10::optional with std::optional in topk kernel (#34467)

Pushpinder Singh committed 9 months ago

[Feature] Support CPU Offloading without Pytorch Pinned Memory that leads to doubled allocation (#32993)

Wei Zhao committed 9 months ago

[Bugfix] Add quant_config in ViT of Kimi-K2.5 (#34501)

LoganJane committed 9 months ago

[Bugfix] Exclude `language_model_only` key from MM AOT compile hash but include in model one (#34508)

Roger Wang committed 9 months ago

[Misc] Port Qwen3.5 Configs (#34512)

Roger Wang committed 9 months ago

Extend ColBERT support to non-standard BERT backbones (#34170)

Ilya Boytsov committed 9 months ago

[GDN] Use CPU tensors to build GDN metadata (#34498)

Woosuk Kwon committed 9 months ago