Method get_answer

startup/RRAG.py:237–273

(self, question, search_chain_length, split_flag=True)

        self.documents = []

    def get_answer(self, question, search_chain_length, split_flag=True):
        self.input = f"The question: {question}"
        self.output = ""
        length = 0
        while True:
            if length>=search_chain_length:
                break
            else:
                self.input = self.input + self.output
                conversation = [{"role": "system","content": "You are a helpful assistant"},{"role": "user","content": self.input}]
                outputs = self.llm.chat(
                    messages=conversation,
                    sampling_params=self.sampling_params
                )
                # stream_output = self.output + outputs# 流式显示
                generated_text = outputs[0].outputs[0].text
                mydict=split_response(generated_text)
                if mydict.get('answer'):
                    answer=mydict.get('answer')
                    self.search_chain.append(mydict)
                    length+=1
                    self.output = self.output + generated_text
                    break
                elif mydict.get('query'):
                    if split_flag:
                        GetStepbystepRetrievalv2(mydict, self.retrieved_ids, self.documents, self.config)
                    else:
                        GetStepbystepRetrieval(mydict, self.retrieved_ids, self.documents, self.config)
                    self.search_chain.append(mydict)
                    length+=1
                    self.output = self.output + generated_text +"\n"+ f"The retrieval documents: {mydict['doc']}" +"\n"
        pdb.set_trace()
        self.history_down = []
        self.search_chain = []
        self.retrieved_ids = []
        self.documents = []
        return self.output

def main():
    parser = argparse.ArgumentParser(description="使用vLLM部署Llama模型进行多轮对话")