Function run_benchmark

RealChart2Code_eval/evaluate_task2.py:1122–1312

(args)

def run_benchmark(args):
    
    data_file = Path(args.data_dir)
    results_dir = Path(args.results_dir) / args.generation_model_name
    results_dir.mkdir(parents=True, exist_ok=True)
    
    print(f"🚀 Starting PlotCraft Benchmark Evaluation")
    print(f"Generation Model: {args.generation_model_name}")
    print(f"Evaluation Model: {args.evaluation_model_name}")
    print(f"Data File: {data_file}")
    print(f"Results Directory: {results_dir}")
    print("=" * 60)
    
    
    generation_cache_file = results_dir / "generation_cache.jsonl"
    generation_client = FluxOpenAIChat(
        base_url=args.generation_base_url,
        api_key=args.generation_api_key,
        cache_file=str(generation_cache_file),
        max_retries=args.max_retries,
        max_qps=args.max_qps,
        max_qpm=args.max_qpm,
    )
    
    evaluation_cache_file = results_dir / "evaluation_cache.jsonl"
    evaluation_client = FluxOpenAIChat(
        base_url=args.evaluation_base_url,
        api_key=args.evaluation_api_key,
        cache_file=str(evaluation_cache_file),
        max_retries=args.max_retries,
        max_qps=args.max_qps,
        max_qpm=args.max_qpm,
    )
    
    data_evaluation_cache_file = results_dir / "data_evaluation_cache.jsonl"
    data_evaluation_client = FluxOpenAIChat(
        base_url=args.evaluation_base_url,
        api_key=args.evaluation_api_key,
        cache_file=str(data_evaluation_cache_file),
        max_retries=args.max_retries,
        max_qps=args.max_qps,
        max_qpm=args.max_qpm,
    )
    
    generation_prompt = read_system_prompt(args.generation_prompt_path)
    evaluation_prompt = read_system_prompt(args.evaluation_prompt_path)
    data_evaluation_prompt = read_system_prompt(args.data_evaluation_prompt_path)
    
    print("🔄 Loading task information...")
    task_infos = []
    
    try:
        with open(data_file, 'r', encoding='utf-8') as f:
            data = json.load(f)
        
        # 直接读取JSON中的任务列表并转换路径
        for task_key, task_info in data.items():
            # 转换路径字段为Path对象