script/temp_merge.py

import json
import os

# ================= 配置区域 =================
# 已经把 glm_all 系列的后缀全部改为了 .jsonl
FILE_PAIRS = [
    ("glm_all_overfitting_cleaned_ask.jsonl", "sharegpt_h1_nam_num_nhh_len_hard_rej_think_nobitewis_valid_indexed.json"),
    ("glm_all_overfitting_cleaned_ask_left.jsonl", "sharegpt_h1_nam_num_nhh_len_hard_rej_think_nobitewis_valid_indexed_left.json"),
    ("glm_all_overfitting_cleaned_ask_left2.jsonl", "sharegpt_h1_nam_num_nhh_len_hard_rej_think_nobitewis_valid_indexed_left2.json"),
    ("glm_all_overfitting_cleaned_ask_left3.jsonl", "sharegpt_h1_nam_num_nhh_len_hard_rej_think_nobitewis_valid_indexed_left3.json"),
    ("glm_all_overfitting_cleaned_ask_left4.jsonl", "sharegpt_h1_nam_num_nhh_len_hard_rej_think_nobitewis_valid_indexed_left4.json"),
    ("glm_all_overfitting_cleaned_ask_left5_2w.jsonl", "sharegpt_h1_nam_num_nhh_len_hard_rej_think_nobitewis_valid_left5_2w.json"),
    ("glm_all_overfitting_cleaned_ask_left5_1w.jsonl", "sharegpt_h1_nam_num_nhh_len_hard_rej_think_nobitewis_valid_left5_remaining_1w.json"),
    ("glm_all_overfitting_cleaned_ask_left5_r.jsonl", "sharegpt_h1_nam_num_nhh_len_hard_rej_think_nobitewis_valid_left5_remaining.json")
]

# 如果脚本和文件不在同一个目录，请修改 BASE_DIR（例如 "/root/data/"）
BASE_DIR = "/root/test/weitiao/data_process_bq/data3" 

# 输出的合并文件名
OUTPUT_GLM_JSONL = "M_glm_data3_all.jsonl"
OUTPUT_SHAREGPT_JSON = "M_sharegpt_data3_all.json"

# ================= 主逻辑 =================
def main():
    merged_glm_lines = []
    merged_sharegpt_data = []

    print("开始按照对应关系配对并合并文件...\n")

    for idx, (glm_name, sharegpt_name) in enumerate(FILE_PAIRS, 1):
        glm_path = os.path.join(BASE_DIR, glm_name)
        sharegpt_path = os.path.join(BASE_DIR, sharegpt_name)

        # 检查文件是否存在
        if not os.path.exists(glm_path) or not os.path.exists(sharegpt_path):
            print(f"[错误] 找不到文件，跳过第 {idx} 对: \n  - {glm_name}\n  - {sharegpt_name}\n")
            continue

        # 1. 读取 GLM 文件 (.jsonl 按行读取)
        current_glm_lines = []
        with open(glm_path, 'r', encoding='utf-8') as f:
            for line in f:
                line = line.strip()
                if line: # 过滤掉可能的空行
                    current_glm_lines.append(line)
        len_glm = len(current_glm_lines)
        
        # 2. 读取 ShareGPT 文件 (.json 整体解析)
        with open(sharegpt_path, 'r', encoding='utf-8') as f:
            sharegpt_data = json.load(f)
        len_sg = len(sharegpt_data)

        print(f"[{idx}/8] 正在处理:")
        print(f"  🔴 {glm_name} (有效行数: {len_glm})")
        print(f"  🔵 {sharegpt_name} (数组长度: {len_sg})")

        # 3. 核心：严格校验数据条数是否一一对应
        if len_glm != len_sg:
            print(f"  ❌ [警告] 这对文件条数不一致！停止合并操作，请检查原始数据。\n")
            return  # 发现不一致直接终止程序，避免生成错位数据
        else:
            print(f"  ✅ 条数匹配，已加入合并队列。\n")
            merged_glm_lines.extend(current_glm_lines)
            merged_sharegpt_data.extend(sharegpt_data)

    # 4. 保存合并后的文件
    print("================ 合并完成 ================")
    print(f"所有文件校验通过！总数据量为: {len(merged_glm_lines)} 条")
    
    # 保存 JSONL
    print(f"正在保存 -> {OUTPUT_GLM_JSONL}")
    with open(os.path.join(BASE_DIR, OUTPUT_GLM_JSONL), 'w', encoding='utf-8') as f:
        for line in merged_glm_lines:
            f.write(line + '\n')

    # 保存 JSON
    print(f"正在保存 -> {OUTPUT_SHAREGPT_JSON}")
    with open(os.path.join(BASE_DIR, OUTPUT_SHAREGPT_JSON), 'w', encoding='utf-8') as f:
        json.dump(merged_sharegpt_data, f, ensure_ascii=False, indent=2)

    print("\n🎉 全部处理完毕！两份最终合并文件的顺序和条数已100%一一对应。")

if __name__ == "__main__":
    main()