vllm-omni

Public

vLLM-omni supports multi-modality models inference and serving with non-autoregressive structures and non-textual outputs, extending vLLM beyond traditional text-based, autoregressive generation.