make_turn.py

import json
import copy
import argparse
import os

def split_dialogue_by_assistant(input_path, output_path):
    if not os.path.exists(input_path):
        print(f"❌ 錯誤：找不到輸入檔案 '{input_path}'")
        return

    split_results = []
    original_count = 0

    print(f"🚀 開始處理：{input_path}")

    with open(input_path, 'r', encoding='utf-8') as f:
        for line in f:
            line = line.strip()
            if not line:
                continue
            
            try:
                entry = json.loads(line)
                original_count += 1
                
                messages = entry.pop("messages", [])
                metadata = entry 
                
                current_history = []
                assistant_turn_count = 0 # 用來計算這是第幾個 assistant 回覆

                for msg in messages:
                    current_history.append(msg)
                    
                    if msg.get("role") == "assistant":
                        assistant_turn_count += 1
                        
                        # 複製 metadata
                        new_sample = copy.deepcopy(metadata)
                        
                        # --- 修正 ID 邏輯 ---
                        if "id" in new_sample:
                            # 在原始 ID 後面加上序號，例如 "chat_001" 變成 "chat_001_turn_1"
                            new_sample["id"] = f"{new_sample['id']}_turn_{assistant_turn_count}"
                        # -------------------

                        new_sample["messages"] = copy.deepcopy(current_history)
                        split_results.append(new_sample)
            
            except json.JSONDecodeError:
                continue

    with open(output_path, 'w', encoding='utf-8') as f:
        for item in split_results:
            f.write(json.dumps(item, ensure_ascii=False) + '\n')

    print("-" * 30)
    print(f"✅ 處理完成！")
    print(f"📦 原始樣本數：{original_count}")
    print(f"多輪切分後總樣本數：{len(split_results)}")
    print(f"🆔 已自動為 ID 加上 _turn_N 後綴避免重複。")

if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument("--input", "-i", type=str, required=True)
    parser.add_argument("--output", "-o", type=str, default="output_split.jsonl")
    args = parser.parse_args()
    split_dialogue_by_assistant(args.input, args.output)