vllm

[CI] Heavy refactoring of Voxtral multimodal audio model tests

#34294

Merged