| 777 | |
| 778 | @staticmethod |
| 779 | def function_calling( |
| 780 | model_name, |
| 781 | prompt, |
| 782 | tools, |
| 783 | timeout=30, |
| 784 | mode="fast_first", |
| 785 | input_proportion=60, |
| 786 | output_proportion=40, |
| 787 | max_tokens=None, |
| 788 | test_response = None, |
| 789 | remark = "" |
| 790 | ): |
| 791 | # Validate model is allowed for function calling |
| 792 | LLM_Wrapper._validate_model_for_function_calling(model_name) |
| 793 | |
| 794 | if test_response is not None: |
| 795 | return test_response |
| 796 | """ |
| 797 | 使用功能调用能力生成响应 |
| 798 | |
| 799 | Args: |
| 800 | model_name (str): 模型名称 |
| 801 | prompt (str): 输入提示 |
| 802 | tools (list): 工具定义列表 |
| 803 | timeout (int): 请求超时时间（秒） |
| 804 | mode (str): 选择策略，"cheap_first"或"fast_first" |
| 805 | input_proportion (int): 输入比例 |
| 806 | output_proportion (int): 输出比例 |
| 807 | max_tokens (int, optional): 最大生成token数，默认None（不限制） |
| 808 | |
| 809 | Returns: |
| 810 | dict: 包含响应内容和工具调用的完整响应 |
| 811 | |
| 812 | Raises: |
| 813 | ValueError: 当模型在源上不可用时 |
| 814 | Exception: 请求失败或其他错误 |
| 815 | """ |
| 816 | try: |
| 817 | |
| 818 | load_balancing = LoadBalancing() |
| 819 | ( |
| 820 | main_source_name, |
| 821 | main_source_model_name, |
| 822 | main_api_key, |
| 823 | backup_source_name, |
| 824 | backup_source_model_name, |
| 825 | backup_api_key, |
| 826 | ) = load_balancing.get_config( |
| 827 | model_name, mode, input_proportion, output_proportion |
| 828 | ) |
| 829 | |
| 830 | # Verify model mappings are valid |
| 831 | LLM_Wrapper._verify_model_mapping( |
| 832 | main_source_name, |
| 833 | backup_source_name, |
| 834 | main_source_model_name, |
| 835 | backup_source_model_name, |
| 836 | model_name, |