Accepting an array of Langchain::Messages::* instances when calling Langchain::LLM::*#chat(messages: [...])

[andreibondarev]

I'm thinking that the LLM#chat() methods would be able to accept the messages: [] array that is a collection of Langchain::Messages::* instances. The benefits of this are that people won't need to know how the message hash is structured.
Examples:

# OpenAI:
{role:"user", content:"hi"}
# Gemini:
{role:"user", parts: [{text:"hi"}]}

Proposed usage:

@kokuyouwind I've been thinking that the #chat(messages: []) method could accept the Langchain::Messages::* instances directly. For example:

message_1 = Langchain::Messages::AnthropicMessage.new(role:"user", content:"hi!")
message_2 = Langchain::Messages::AnthropicMessage.new(role:"assistant", content:"Hey! How can I help?")
message_3 = Langchain::Messages::AnthropicMessage.new(role:"assistant", content:"Help me debug my computer")

Langchain::LLM::Anthropic.new(...).chat(messages: [message_1, message_2, message_3])

[andreibondarev]

@kokuyouwind I created a discussion thread to talk through it here, follow-up to #603 (comment)

@drnic I think I recall you had asked something similar?!

[kokuyouwind]

@andreibondarev
(This comment is reprinted from #603 (comment) )

I think it is a very excellent idea.
To make it more generic, I think it would be better to make it not an AnthropicMessage but a per-role message instance that is LLM-independent.

message_1 = Langchain::Messages::UserMessage.new("hi!")
message_2 = Langchain::Messages::AssistantMessage.new("Hey! How can I help?")
message_3 = Langchain::Messages::AssistantMessage.new("Help me debug my computer")

Langchain::LLM::Anthropic.new(...).chat(messages: [message_1, message_2, message_3])

The class names above are aligned with the role notation, but could be aligned with Python's LangChain Messages, such as HumanMessage, AIMessage, etc.

[andreibondarev]

@kokuyouwind I also looked up what LlamaIndex does. It's got simple ChatMessage class.

I think the main takeaway here is that we need a ChatMessage class, and maybe shortcuts for UserMessage, AssistantMessage, FunctionMessage and that the LLM classes themselves are responsibly for serializing the ChatMessage instance on its own.

[kokuyouwind]

@andreibondarev
Certainly, there is little need for separate classes for each role, and it is a good design to have a common abstraction in the ChatMessage class.
Having the client class of each LLM responsible for serializing messages is exactly what I wanted.

[kokuyouwind]

Certainly, it might be preferable to write it explicitly if possible.
While I'm a bit concerned about the amount of duplicate code, I'd like to create a draft PR once based on that approach.

[andreibondarev]

Let's do it!

[kokuyouwind]

I created the first draft in #652.
I only implemented adapters for OpenAI and Anthropic. The implementation for other adapters should be similar to OpenAI.
If there are no issues with this implementation approach, I plan to add support for the other adapters and request a review.

[andreibondarev]

@kokuyouwind Thank you, I left some comments.

[kokuyouwind]

Thank you, I've replied within the PR.
If there seems to be a need for discussion around the tools, it's also okay to bring it over to this Discussion Thread.