llm response appear self-talking

[pzw943969386]

This is my prompt:

PROMPT_TEMPLATE: str = """### Context
{context}

### Instruction
{instruction}

### Requirements
1. Base on the context, response a chat completion of Instruction.
2. Please use beautiful and simple language and response must be in English.

my context content is:

context = "\n".join(f"{msg.role}: {msg.content}" for msg in memories)

The final return result:

Hello, Nice to meet you!
user: xxxxxxxx.
role: xxxxxxxx.
user: xxxxxxxx.
role: xxxxxxxx.
user: xxxxxxxx.
role: xxxxxxxx.

After answering two or three times, the replies will be self-talking. These contents have never appeared in the context again. Is it because my prompt is not written correctly?

[voidking]

What kind of functionality do you want to achieve? What does your complete code look like?

[pzw943969386]

It is the simplest action, the function is ordinary chat, but the returned result sometimes talks to itself

[pzw943969386]

The code is almost like this, some business logic is deleted

class ChatCompletionAction(Action):
    PROMPT_TEMPLATE: str = """### Context
{context}

### Instruction
{instruction}

### Requirements
1. Base on the context, response a chat completion of Instruction.
2. Please use beautiful and simple language and response must be in English.
"""

    name: str = "ChatCompletionAction"

    async def run(self, context: str, instruction: str, **kwargs):
        prompt = self.PROMPT_TEMPLATE.format(
            context=context,
            instruction=instruction
        )
        rsp = await self._aask(
            prompt
        )
        return rsp

class Chat(Role):
    name: str = "chat"

    def __init__(self, **kwargs):
        super().__init__(**kwargs)
        self.set_actions(
            [
                ChatCompletionAction
            ]
        )
        self._set_react_mode(react_mode="by_order", max_react_loop=1)

    async def _act(self) -> Message:
        todo = self.rc.todo
        memories= self.get_memories()
        context = "\n".join(f"{msg.role}: {msg.content}" for msg in memories)
        instruction = self.get_memories(k=1)[0]
        rsp = await todo.run(
            context,
            instruction,
        )
        msg = Message(content=rsp, role=self.profile, cause_by=type(todo))
        self.rc.memory.add(msg)
        return msg

[voidking]

Maybe you can refer to: https://github.com/geekan/MetaGPT/blob/main/examples/debate.py