---
license: apache-2.0
datasets:
- ShengbinYue/DISC-Law-SFT
base_model:
- Qwen/Qwen2.5-Coder-7B
library_name: adapter-transformers
tags:
- legal
---
# Qwen2.5-Coder-7B-LoRA-Law

This repository contains trained weights for a fine-tuned version of the Qwen2.5-Coder-7B-Instruct model, specialized for answering legal questions in Chinese.  The model is fine-tuned using LoRA on the DISC-Law-SFT dataset.

## Model Description

This model is intended to provide informative and comprehensive answers to legal questions posed in Chinese.  

## Training Details

* **Base Model:** Qwen/Qwen2.5-Coder-7B-Instruct
* **Dataset:** [DISC-Law-SFT](https://huggingface.co/datasets/ShengbinYue/DISC-Law-SFT), specifically the `DISC-Law-SFT-Pair-QA-released.jsonl` subset.
* **Fine-tuning Method:** LoRA
* **LoRA Parameters:**
    * `r`: 64
    * `lora_alpha`: 16
    * `lora_dropout`: 0.1
* **Training Hyperparameters:**
    * `per_device_train_batch_size`: 4
    * `gradient_accumulation_steps`: 8
    * `learning_rate`: 2e-5
    * `num_train_epochs`: 2
* **System Prompt:** 你是一个法律专家，请根据用户的问题给出专业的回答


## How to Merge Lora weights


```python
from peft import PeftModel
from transformers import AutoModelForCausalLM
import torch
import os
from modelscope import AutoTokenizer
import shutil

# 保证原始模型的各个文件不遗漏保存到merge_path中
def copy_files_not_in_B(A_path, B_path):
    """
    Copies files from directory A to directory B if they exist in A but not in B.

    :param A_path: Path to the source directory (A).
    :param B_path: Path to the destination directory (B).
    """
    # 保证路径存在
    if not os.path.exists(A_path):
        raise FileNotFoundError(f"The directory {A_path} does not exist.")
    if not os.path.exists(B_path):
        os.makedirs(B_path)

    # 获取路径A中所有非权重文件
    files_in_A = os.listdir(A_path)
    files_in_A = set([file for file in files_in_A if not (".bin" in file or "safetensors" in file)])
    # List all files in directory B
    files_in_B = set(os.listdir(B_path))

    # 找到所有A中存在但B中不存在的文件
    files_to_copy = files_in_A - files_in_B

    # 将文件或文件夹复制到B路径下
    for file in files_to_copy:
        src_path = os.path.join(A_path, file)
        dst_path = os.path.join(B_path, file)

        if os.path.isdir(src_path):
            # 复制目录及其内容
            shutil.copytree(src_path, dst_path)
        else:
            # 复制文件
            shutil.copy2(src_path, dst_path)

def merge_lora_to_base_model():
    model_name_or_path = '...'  # 原模型地址
    adapter_name_or_path = '...'  # 微调后模型的保存地址
    save_path = '...'

    # 如果文件夹不存在，就创建
    if not os.path.exists(save_path):
        os.makedirs(save_path)
    tokenizer = AutoTokenizer.from_pretrained(model_name_or_path,trust_remote_code=True,)

    model = AutoModelForCausalLM.from_pretrained(
        model_name_or_path,
        trust_remote_code=True,
        low_cpu_mem_usage=True,
        torch_dtype=torch.float16,
        device_map="auto"
    )
    # 加载保存的 Adapter
    model = PeftModel.from_pretrained(model, adapter_name_or_path, device_map="auto",trust_remote_code=True)
    # 将 Adapter 合并到基础模型中
    merged_model = model.merge_and_unload()  # PEFT 的方法将 Adapter 权重合并到基础模型
    # 保存合并后的模型
    tokenizer.save_pretrained(save_path)
    merged_model.save_pretrained(save_path, safe_serialization=False)
    copy_files_not_in_B(model_name_or_path, save_path)
    print(f"合并后的模型已保存至: {save_path}")

if __name__ == '__main__':
    merge_lora_to_base_model()