Spaces:
Runtime error
Runtime error
| # Copyright (c) Microsoft. All rights reserved. | |
| from typing import Optional | |
| import semantic_kernel, autogen | |
| class AutoGenPlanner: | |
| """(Demo) Semantic Kernel planner using Conversational Programming via AutoGen. | |
| AutoGenPlanner leverages OpenAI Function Calling and AutoGen agents to solve | |
| a task using only the Plugins loaded into Semantic Kernel. SK Plugins are | |
| automatically shared with AutoGen, so you only need to load the Plugins in SK | |
| with the usual `kernel.import_skill(...)` syntax. You can use native and | |
| semantic functions without any additional configuration. Currently the integration | |
| is limited to functions with a single string parameter. The planner has been | |
| tested with GPT 3.5 Turbo and GPT 4. It always used 3.5 Turbo with OpenAI, | |
| just for performance and cost reasons. | |
| """ | |
| import datetime | |
| from typing import List, Dict | |
| ASSISTANT_PERSONA = f"""Only use the functions you have been provided with. | |
| Do not ask the user to perform other actions than executing the functions. | |
| Use the functions you have to find information not available. | |
| Today's date is: {datetime.date.today().strftime("%B %d, %Y")}. | |
| Reply TERMINATE when the task is done. | |
| """ | |
| def __init__(self, kernel: semantic_kernel.Kernel, llm_config: Dict = None): | |
| """ | |
| Args: | |
| kernel: an instance of Semantic Kernel, with plugins loaded. | |
| llm_config: a dictionary with the following keys: | |
| - type: "openai" or "azure" | |
| - openai_api_key: OpenAI API key | |
| - azure_api_key: Azure API key | |
| - azure_deployment: Azure deployment name | |
| - azure_endpoint: Azure endpoint | |
| """ | |
| super().__init__() | |
| self.kernel = kernel | |
| self.llm_config = llm_config | |
| def create_assistant_agent(self, name: str, persona: str = ASSISTANT_PERSONA) -> autogen.AssistantAgent: | |
| """ | |
| Create a new AutoGen Assistant Agent. | |
| Args: | |
| name (str): the name of the agent | |
| persona (str): the LLM system message defining the agent persona, | |
| in case you want to customize it. | |
| """ | |
| return autogen.AssistantAgent(name=name, system_message=persona, llm_config=self.__get_autogen_config()) | |
| def create_user_agent( | |
| self, name: str, max_auto_reply: Optional[int] = None, human_input: Optional[str] = "ALWAYS" | |
| ) -> autogen.UserProxyAgent: | |
| """ | |
| Create a new AutoGen User Proxy Agent. | |
| Args: | |
| name (str): the name of the agent | |
| max_auto_reply (int): the maximum number of consecutive auto replies. | |
| default to None (no limit provided). | |
| human_input (str): the human input mode. default to "ALWAYS". | |
| Possible values are "ALWAYS", "TERMINATE", "NEVER". | |
| (1) When "ALWAYS", the agent prompts for human input every time a message is received. | |
| Under this mode, the conversation stops when the human input is "exit", | |
| or when is_termination_msg is True and there is no human input. | |
| (2) When "TERMINATE", the agent only prompts for human input only when a termination message is received or | |
| the number of auto reply reaches the max_consecutive_auto_reply. | |
| (3) When "NEVER", the agent will never prompt for human input. Under this mode, the conversation stops | |
| when the number of auto reply reaches the max_consecutive_auto_reply or when is_termination_msg is True. | |
| """ | |
| return autogen.UserProxyAgent( | |
| name=name, | |
| human_input_mode=human_input, | |
| max_consecutive_auto_reply=max_auto_reply, | |
| function_map=self.__get_function_map(), | |
| ) | |
| def __get_autogen_config(self): | |
| """ | |
| Get the AutoGen LLM and Function Calling configuration. | |
| """ | |
| if self.llm_config: | |
| if self.llm_config["type"] == "openai": | |
| if not self.llm_config["openai_api_key"] or self.llm_config["openai_api_key"] == "sk-...": | |
| raise Exception("OpenAI API key is not set") | |
| return { | |
| "functions": self.__get_function_definitions(), | |
| "config_list": [{"model": "gpt-3.5-turbo", "api_key": self.llm_config["openai_api_key"]}], | |
| } | |
| if self.llm_config["type"] == "azure": | |
| if ( | |
| not self.llm_config["azure_api_key"] | |
| or not self.llm_config["azure_deployment"] | |
| or not self.llm_config["azure_endpoint"] | |
| ): | |
| raise Exception("Azure OpenAI API configuration is incomplete") | |
| return { | |
| "functions": self.__get_function_definitions(), | |
| "config_list": [ | |
| { | |
| "model": self.llm_config["azure_deployment"], | |
| "api_type": "azure", | |
| "api_key": self.llm_config["azure_api_key"], | |
| "api_base": self.llm_config["azure_endpoint"], | |
| "api_version": "2023-08-01-preview", | |
| } | |
| ], | |
| } | |
| raise Exception("LLM type not provided, must be 'openai' or 'azure'") | |
| def __get_function_definitions(self) -> List: | |
| """ | |
| Get the list of function definitions for OpenAI Function Calling. | |
| """ | |
| functions = [] | |
| sk_functions = self.kernel.skills.get_functions_view() | |
| for ns in {**sk_functions.native_functions, **sk_functions.semantic_functions}: | |
| for f in sk_functions.native_functions[ns]: | |
| functions.append( | |
| { | |
| "name": f.name, | |
| "description": f.description, | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| f.parameters[0].name: { | |
| "description": f.parameters[0].description, | |
| "type": f.parameters[0].type_, | |
| } | |
| }, | |
| "required": [f.parameters[0].name], | |
| }, | |
| } | |
| ) | |
| return functions | |
| def __get_function_map(self) -> Dict: | |
| """ | |
| Get the function map for AutoGen Function Calling. | |
| """ | |
| function_map = {} | |
| sk_functions = self.kernel.skills.get_functions_view() | |
| for ns in {**sk_functions.native_functions, **sk_functions.semantic_functions}: | |
| for f in sk_functions.native_functions[ns]: | |
| function_map[f.name] = self.kernel.skills.get_function(f.skill_name, f.name) | |
| return function_map | |