Method run

demo/vimrag_agent.py:339–453

Run the VimRAG agent on a sample query. Yields progress events in real-time. Args: sample: Dict with 'query' key Yields: Dict with 'event' key and event-specific data: - {"event": "think", "content": str} - Thinking c

(self, sample)

            multimodal_memory[last_graph_node['id']] = node_memory

    def run(self, sample):
        """
        Run the VimRAG agent on a sample query. Yields progress events in real-time.
        
        Args:
            sample: Dict with 'query' key
            
        Yields:
            Dict with 'event' key and event-specific data:
            - {"event": "think", "content": str} - Thinking content chunk
            - {"event": "content", "content": str} - Output content chunk
            - {"event": "search", "query": str} - Search initiated
            - {"event": "search_done", "results": dict} - Search completed
            - {"event": "memorize", "summary": str} - Memorize action
            - {"event": "answer", "content": str, "sample": dict} - Final answer
            - {"event": "error", "content": str} - Error occurred
            - {"event": "max_steps", "content": str} - Maximum steps reached
        """
        question = sample['query']
        trajectory = []
        search_results_list = []
        
        # Initialize action graph with root node
        action_graph = [{
            "id": "root",
            "name": "Initial Node",
            "content": f"Initial query from user: {question}"
        }]
        multimodal_memory = {}
        
        need_update_context = True
        can_search = True
        last_graph_node = None
        vision_ids_dict = None
        generate_times = 0
        steps_remaining = self.max_mem_steps

        while steps_remaining > 0:
            steps_remaining -= 1
            
            # Build or update context
            if need_update_context:
                messages = self._build_initial_messages(question, action_graph)
                self._update_messages_with_memory(messages, action_graph, multimodal_memory)
                need_update_context = False
                can_search = True

            try:
                # Generate model response with streaming
                messages_base64 = fast_process_messages(messages)
                
                # 流式收集模型输出
                full_response = ""
                full_reasoning = ""
                for chunk in self._model_generate(messages_base64):
                    if chunk["type"] == "think":
                        yield {"event": "think", "content": chunk["content"]}
                    elif chunk["type"] == "content":