auto-round
fix: use spawn start method for vLLM evaluation to avoid CUDA re-initialization error
#2421
Merged

Loading