cd /home/mshahidul/readctrl/code/RL_model/verl/verl_train
python scripts/legacy_model_merger.py merge \
    --backend fsdp \
    --local_dir /home/mshahidul/readctrl/code/RL_model/RL_model_subclaim_classifier/global_step_45/actor \
    --target_dir /home/mshahidul/readctrl/code/RL_model/converted_model/v1


CUDA_DEVICE_ORDER=PCI_BUS_ID CUDA_VISIBLE_DEVICES=2 python -m vllm.entrypoints.openai.api_server \
  --model /home/mshahidul/readctrl/code/RL_model/converted_model/v1 \
  --served-model-name inference \
  --dtype bfloat16 \
  --port 8001

python /home/mshahidul/readctrl/code/RL_model/run_inference_vllm_server.py \
  --base_url http://127.0.0.1:8001/v1 \
  --served_model_name inference \
  --batch_size 8