import json
import numpy as np
import os
from peft import PeftModel
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig

def merge(base_model_path, peft_model_path, save_path):
    base_model = AutoModelForCausalLM.from_pretrained(base_model_path, torch_dtype="bfloat16")
    ft_model = PeftModel.from_pretrained(base_model, peft_model_path)
    ft_model = ft_model.merge_and_unload()
    ft_model.save_pretrained(save_path)

def main():
    file_path = os.path.dirname(__file__)

    base_model_path = file_path + "/../../data/models/Qwen3-Embedding-0.6B/model"
    peft_model_path = file_path + "/output/v1-20251122-184545/checkpoint-3434"
    save_path = file_path + "/output/v1-20251122-184545/merged_checkpoint-3434"
    merge(base_model_path, peft_model_path, save_path)


if __name__ == "__main__":
    main()