Update run_checkpoints.sh to use the vllm api server instead of offline inference

Update run_checkpoints.sh to use the vllm api server instead of offline inference #387