LLMConversation

LLMConversation is an abstract base class designed to represent different types of LLM (Language Learning Module) conversations. It is a part of the library, automata.llm.foundation.

Overview

LLMConversation is an outline or blueprint for implementing different conversation types. As an abstract base class, it lays out a set of methods that should be implemented in the child classes. Some of the methods specified by this class involve getting the latest message, registering observers, and managing messages within a conversation.

One of the exceptions managed by this module is the LLMEmptyConversationError, raised when an operation is invoked for an empty conversation.

Example

Below is a sample implementation of LLMConversation and usage of LLMEmptyConversationError. As LLMConversation is an abstract base class, we need to first implement all abstract methods in a subclass:

from automata.llm.foundation.LLMConversation import LLMEmptyConversationError, LLMChatMessage
from automata.core.base.patterns.observer import Observer

class CustomConversation(LLMConversation):
    def __init__(self):
        self._observers: Set[Observer] = set()
        self.messages: List[LLMChatMessage] = []

    def get_latest_message(self) -> LLMChatMessage:
        try:
            return self.messages[-1]
        except IndexError:
            raise LLMEmptyConversationError()

    # Note: Similarly implement the rest of the abstract methods here...

# Usage
conversation = CustomConversation()

try:
    latest_message = conversation.get_latest_message()
except LLMEmptyConversationError:
    print("Conversation is currently empty.")

Limitations

This class is an abstract base class, so it cannot be instantiated directly. Additionally, it does not provide default implementations for its abstract methods requiring subclasses to provide their own specific implementations. Also, error handling for empty conversations is a responsibility of the class’s consumers.

Follow-up Questions:

Are there any specifications on how the other abstract methods (like __len__, get_messages_for_next_completion, and notify_observers) should be implemented?
How are observers expected to interact with this class?