Function parse_args

scripts/run_web_thinker_report.py:92–126

()

]

def parse_args():
    parser = argparse.ArgumentParser(description="Run Search-o1 for various datasets and models.")
    parser.add_argument('--single_question', type=str, default=None, help="Single question to process instead of dataset")
    parser.add_argument('--dataset_name', type=str, required=False, default='custom', help="Name of the dataset to use.")
    parser.add_argument('--split', type=str, required=False, default='test', help="Dataset split to use.")
    parser.add_argument('--subset_num', type=int, default=-1, help="Number of examples to process. Defaults to all if not specified.")

    parser.add_argument('--temperature', type=float, default=0.7, help="Sampling temperature.")
    parser.add_argument('--top_p', type=float, default=0.8, help="Top-p sampling parameter.")
    parser.add_argument('--min_p', type=float, default=0.05, help="Minimum p sampling parameter.")
    parser.add_argument('--top_k_sampling', type=int, default=20, help="Top-k sampling parameter.")
    parser.add_argument('--repetition_penalty', type=float, default=1.05, help="Repetition penalty. If not set, defaults based on the model.")
    parser.add_argument('--max_tokens', type=int, default=81920, help="Maximum number of tokens to generate. If not set, defaults based on the model and dataset.")

    # parser.add_argument('--max_search_limit', type=int, default=10, help="Maximum number of searches per question.")
    parser.add_argument('--top_k', type=int, default=10, help="Maximum number of search documents to return.")
    parser.add_argument('--keep_links', action='store_true', default=False, help="Whether to keep links in fetched web content")
    parser.add_argument('--use_jina', action='store_true', help="Whether to use Jina API for document fetching.")
    parser.add_argument('--jina_api_key', type=str, default='None', help="Your Jina API Key to Fetch URL Content.")
    parser.add_argument('--bing_subscription_key', type=str, default=None, help="Bing Search API subscription key.")
    parser.add_argument('--bing_endpoint', type=str, default="https://api.bing.microsoft.com/v7.0/search", help="Bing Search API endpoint.")
    parser.add_argument('--serper_api_key', type=str, default=None, help="Google Serper API key.")
    parser.add_argument('--search_engine', type=str, default="bing", choices=["bing", "serper"], help="Search engine to use (bing or serper). Default: bing")
    parser.add_argument('--eval', action='store_true', help="Whether to run evaluation after generation.")
    parser.add_argument('--seed', type=int, default=None, help="Random seed for generation. If not set, will use current timestamp as seed.")
    parser.add_argument('--api_base_url', type=str, required=True, help="Base URL for the API endpoint")
    parser.add_argument('--aux_api_base_url', type=str, required=True, help="Base URL for the auxiliary model API endpoint")
    parser.add_argument('--model_name', type=str, default="QwQ-32B", help="Name of the model to use")
    parser.add_argument('--aux_model_name', type=str, default="Qwen2.5-32B-Instruct", help="Name of the auxiliary model to use")
    parser.add_argument('--concurrent_limit', type=int, default=32, help="Maximum number of concurrent API calls")
    parser.add_argument('--lora_name', type=str, default=None, help="Name of the LoRA adapter to load")
    parser.add_argument('--lora_path', type=str, default=None, help="Path to the LoRA weights")
    parser.add_argument('--tokenizer_path', type=str, default="/share/project/llm/QwQ-32B", help="Path to the main tokenizer")
    parser.add_argument('--aux_tokenizer_path', type=str, default="/share/project/llm/Qwen2.5-32B-Instruct", help="Path to the auxiliary tokenizer")
    return parser.parse_args()

# Initialize tokenizers
args = parse_args()