Class advancedMemAgent

tasks/AutoMem/code/eval.py:51–209

sentence_model = None

class advancedMemAgent:
    def __init__(self, model, backend, retrieve_k, temperature_c5, sglang_host="http://localhost", sglang_port=30000):
        self.memory_system = AgenticMemorySystem(
            model_name='all-MiniLM-L6-v2',
            llm_backend=backend,
            llm_model=model,
            sglang_host=sglang_host,
            sglang_port=sglang_port
        )
        self.retriever_llm = LLMController(
            backend=backend,
            model=model,
            api_key=None,
            sglang_host=sglang_host,
            sglang_port=sglang_port
        )
        self.retrieve_k = retrieve_k
        self.temperature_c5 = temperature_c5

    def add_memory(self, content, time=None):
        self.memory_system.add_note(content, time=time)

    def retrieve_memory(self, content, k=10):
        return self.memory_system.find_related_memories_raw(content, k=k)

    def retrieve_memory_llm(self, memories_text, query):
        prompt = f"""Given the following conversation memories and a question, select the most relevant parts of the conversation that would help answer the question. Include the date/time if available.

                Conversation memories:
                {memories_text}

                Question: {query}

                Return only the relevant parts of the conversation that would help answer this specific question. Format your response as a JSON object with a "relevant_parts" field containing the selected text.
                If no parts are relevant, do not do any things just return the input.

                Example response format:
                {{"relevant_parts": "2024-01-01: Speaker A said something relevant..."}}"""

            # Get LLM response
        response = self.retriever_llm.llm.get_completion(prompt,response_format={"type": "json_schema", "json_schema": {
                            "name": "response",
                            "schema": {
                                "type": "object",
                                "properties": {
                                    "relevant_parts": {
                                        "type": "string",
                                    }
                                },
                                "required": ["relevant_parts"],
                                "additionalProperties": False
                            },
                            "strict": True
                        }})
        # print("response:{}".format(response))
        return response

    def generate_query_llm(self, question):