search

langchain-ai / langchain

🦜🔗 Build context-aware reasoning applications
https://python.langchain.com
MIT License
89.23k stars 14.06k forks source link

Unable to reuse Chat Models for Anthropic Claude #2638

Closed samching closed 5 months ago

samching commented 1 year ago

Problem

Langchain currently doesn't support chat format for Anthropic (e.g. being able to use HumanMessage and AIMessage classes)

Currently, when testing the same prompt across both Anthropic and OpenAI chat models, it requires rewriting the same prompt, although they fundamentally use the same Human:... AI:.... structure.

This means duplicating 2 * n chains prompts (and more if you write separate prompts for turbo-3.5 and 4 (likewise for instant and v1.2 for Claude), making it very unwieldy to test and scale the number of chains.

Potential Solution

  1. Create a wrapper class ChatClaude and add a function like this to translate both AIMessage and HumanMessage to anthropic.AI_PROMPT and anthropic.HUMAN_PROMPT respectively.

But, definitely also open to other solutions which could work here.

supreetkt commented 1 year ago

Have you tried ChatAnthropic?

evolate-tim commented 9 months ago

Working with ConversationBufferWindowMemory and anthropic LLM's will also throw a lot of warnings since the Base template "_DEFAULT_SUMMARIZER_TEMPLATE" has "AI:" examples in it. The summarization will however work in the end.

dosubot[bot] commented 5 months ago

Hi, @samching,

I'm helping the LangChain team manage their backlog and am marking this issue as stale. From what I understand, you raised a concern about Langchain not supporting chat format for Anthropic, and the proposed solution was to create a wrapper class and add a function to translate AIMessage and HumanMessage to anthropic.AI_PROMPT and anthropic.HUMAN_PROMPT. In the comments, there were suggestions to try ChatAnthropic, and warnings were mentioned when working with ConversationBufferWindowMemory and anthropic LLM's due to examples in the Base template.

Could you please confirm if this issue is still relevant to the latest version of the LangChain repository? If it is, please let the LangChain team know by commenting on the issue. Otherwise, feel free to close the issue yourself, or it will be automatically closed in 7 days.

Thank you for your understanding and cooperation.