Please refer to the examples in examples/infer and change swift infer to swift deploy to start the service. (You need to additionally remove --val_dataset)
e.g.
CUDA_VISIBLE_DEVICES=0 \
swift deploy \
--model Qwen/Qwen2.5-7B-Instruct \
--infer_backend vllm