scripts/run_llm_cot.py

# scripts/run_llm_cot.py

import argparse
import json
import os
import sys
from dotenv import load_dotenv

# 加载环境变量
load_dotenv()

sys.path.append(os.path.join(os.path.dirname(__file__), '..'))

from src.llm_generation.api_client import AIAPIClient # 注意类名变化
from src.llm_generation.generator import CoTGenerator

def load_jsonl(path):
    data = []
    with open(path, 'r') as f:
        for line in f:
            data.append(json.loads(line))
    return data
def save_jsonl(data, path):
    with open(path, 'w', encoding='utf-8') as f:
        for item in data:
            f.write(json.dumps(item, ensure_ascii=False) + '\n')

def main():
    parser = argparse.ArgumentParser()
    parser.add_argument("--input_file", type=str, required=True)
    parser.add_argument("--output_file", type=str, required=True)
    parser.add_argument("--image_root", type=str, required=True, help="Root directory for images (e.g., /path/to/project)")
    parser.add_argument("--model", type=str, default="gpt-5")
    # 允许命令行覆盖环境变量
    parser.add_argument("--api_key", type=str, default=os.getenv("API_KEY")) 
    parser.add_argument("--base_url", type=str, default=os.getenv("BASE_URL"))
    args = parser.parse_args()

    # 检查 URL 格式
    base_url = args.base_url
    if base_url and not base_url.endswith("/chat/completions"):
        # 简单的自动修正，假设是标准 OpenAI 格式
        if not base_url.endswith("/v1"):
             base_url = f"{base_url.rstrip('/')}/v1/chat/completions"
        else:
             base_url = f"{base_url.rstrip('/')}/chat/completions"
        print(f"⚠️ Auto-corrected BASE_URL to: {base_url}")

    print(f"Loading oracle data from {args.input_file}...")
    oracle_data = load_jsonl(args.input_file)
    
    # 初始化 Client
    client = AIAPIClient(
        api_key=args.api_key,
        base_url=base_url,
        api_type="openai"
    )
    
    # 初始化 Generator
    generator = CoTGenerator(
        client,
        image_root=args.image_root,
        model_name=args.model
    )

    print("Starting CoT generation...")
    final_data = generator.process_batch(oracle_data)

    print(f"Saving {len(final_data)} entries to {args.output_file}...")
    save_jsonl(final_data, args.output_file)
    print("Done!")

if __name__ == "__main__":
    main()