matrix-pygmalion-bot/matrix_pygmalion_bot/bot/ai/langchain_memory.py

from typing import Any, Dict, List

from langchain.chains.llm import LLMChain
from langchain.memory.chat_memory import BaseChatMemory
from langchain.memory.prompt import SUMMARY_PROMPT
from langchain.prompts.base import BasePromptTemplate
from langchain.schema import BaseLanguageModel, BaseMessage, get_buffer_string

from ..utilities.messages import Message

class ChatMessageHistory(BaseModel):
    messages: List[Message] = []

    def add_user_message(self, message: Message) -> None:
        self.messages.append(message)

    def add_ai_message(self, message: Message) -> None:
        self.messages.append(message)

    def add_system_message(self, message: Message) -> None:
        self.messages.append(message)

    def add_chat_message(self, message: Message) -> None:
        self.messages.append(message)

    def clear(self) -> None:
        self.messages = []


class TestMemory(BaseMemory):
    """Buffer for storing conversation memory."""

    human_prefix: str = "Human"
    ai_prefix: str = "AI"
    chat_memory: ChatMessageHistory = Field(default_factory=ChatMessageHistory)
#    buffer: str = ""
    output_key: Optional[str] = None
    input_key: Optional[str] = None
    memory_key: str = "history"  #: :meta private:


    @property
    def memory_variables(self) -> List[str]:
        """Will always return list of memory variables.
        :meta private:
        """
        return [self.memory_key]

    def load_memory_variables(self, inputs: Dict[str, Any]) -> Dict[str, Any]:
        """Return history buffer."""
        string_messages = []
        for m in chat_memory.messages:
            string_messages.append(f"{message.user_name}: {message.message}")
        return {self.memory_key: "\n".join(string_messages)}

    def save_context(self, inputs: Dict[str, Any], outputs: Dict[str, str]) -> None:
        """Save context from this conversation to buffer."""
        input_str, output_str = self._get_input_output(inputs, outputs)
        self.chat_memory.add_user_message(input_str)
        self.chat_memory.add_ai_message(output_str)

    def clear(self) -> None:
        """Clear memory contents."""
        self.chat_memory.clear()


class BotConversationSummaryBufferWindowMemory(BaseChatMemory):
    """Buffer for storing conversation memory."""

    human_prefix: str = "Human"
    ai_prefix: str = "AI"
    # Define key to pass information about entities into prompt.
    memory_key: str = "history"  #: :meta private:
    #k: int = 5
    max_token_limit: int = 1200
    min_token_limit: int = 200
    moving_summary_buffer: str = ""
    
    llm: BaseLanguageModel
    summary_prompt: BasePromptTemplate = SUMMARY_PROMPT

    @property
    def buffer(self) -> List[BaseMessage]:
        """String buffer of memory."""
        return self.chat_memory.messages

    @property
    def memory_variables(self) -> List[str]:
        """Will always return list of memory variables.
        :meta private:
        """
        return [self.memory_key]

    def load_memory_variables(self, inputs: Dict[str, Any]) -> Dict[str, str]:
        """Return history buffer."""
        buffer = self.buffer
        #buffer: Any = self.buffer[-self.k * 2 :] if self.k > 0 else []
        if not self.return_messages:
            buffer = get_buffer_string(
                buffer,
                human_prefix=self.human_prefix,
                ai_prefix=self.ai_prefix,
            )
        return {self.memory_key: buffer}

    def save_context(self, inputs: Dict[str, Any], outputs: Dict[str, str]) -> None:
        """Save context from this conversation to buffer. Pruned."""
        super().save_context(inputs, outputs)
        # Prune buffer if it exceeds max token limit
        buffer = self.chat_memory.messages
        curr_buffer_length = self.llm.get_num_tokens_from_messages(buffer)
        if curr_buffer_length > self.max_token_limit:
            pruned_memory = []
            while curr_buffer_length > self.min_token_limit:
                pruned_memory.append(buffer.pop(0))
                curr_buffer_length = self.llm.get_num_tokens_from_messages(buffer)
            self.moving_summary_buffer = self.predict_new_summary(
                pruned_memory, self.moving_summary_buffer
            )

    def clear(self) -> None:
        """Clear memory contents."""
        super().clear()
        self.moving_summary_buffer = ""


    def predict_new_summary(
        self, messages: List[BaseMessage], existing_summary: str
    ) -> str:
        new_lines = get_buffer_string(
            messages,
            human_prefix=self.human_prefix,
            ai_prefix=self.ai_prefix,
        )

        chain = LLMChain(llm=self.llm, prompt=self.summary_prompt)
        return chain.predict(summary=existing_summary, new_lines=new_lines)