merge_model.py

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

BASE_MODEL = "Qwen/Qwen2.5-Coder-7B-Instruct"
ADAPTER_DIR = "./qwen-7b-coder-nl2sql-grpo/final"
OUTPUT_DIR = "./qwen-7b-nl2sql-merged"

def main():
    print("Loading Base Model...")
    base_model = AutoModelForCausalLM.from_pretrained(
        BASE_MODEL,
        torch_dtype=torch.bfloat16,
        device_map="auto"
    )

    print("Loading Adapters and Merging...")
    # Load the LoRA adapters into the base model
    model = PeftModel.from_pretrained(base_model, ADAPTER_DIR)
    
    # Merge weights permanently
    merged_model = model.merge_and_unload()

    print("Saving Merged Model...")
    merged_model.save_pretrained(OUTPUT_DIR)
    
    tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)
    tokenizer.save_pretrained(OUTPUT_DIR)
    print(f"Done! Merged model saved to {OUTPUT_DIR}")

if __name__ == "__main__":
    main()