langchain: catch if there are only system messages in a prompt for anthropic

[mathislindner]

PR message: Not sure if I put the check at the right spot, but I thought throwing the error before the loop made sense to me.
Description: Checks if there are only system messages using AnthropicChat model and throws an error if it's the case. Check Issue for more details
Issue: #30764

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
langchain	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Apr 23, 2025 1:27pm

[ccurme]

Instead of blocking this ourselves, could we catch the error from Anthropic and add whatever clarifications we want to the error message? This way if Anthropic adds support we will not be blocking.

Here's an example:

langchain/libs/partners/openai/langchain_openai/chat_models/base.py

Line 383 in 3a64c71

def _handle_openai_bad_request(e: openai.BadRequestError) -> None:

Will need to handle for all invocation modes (sync/async invoke and stream)

[mathislindner]

okay, I just thought it would be wiser to do this before sending a call to Anthropic that would get us a bad request, but yes it does makes sense to expect them to change it down the line :)

[mathislindner]

from langchain_core.messages import SystemMessage, HumanMessage, AIMessage
from langchain.chat_models import init_chat_model
import asyncio

LLM = init_chat_model("claude-3-5-haiku-20241022", model_provider = "anthropic")
prompt = SystemMessage("test")

def test_generate():
    try:
        _ = LLM._generate([prompt])
    except ValueError as e:
        assert str(e) == "Received only system message(s). "
        
def test_stream():
    try:
        stream = LLM._stream([prompt])
        for _ in stream:
            pass
    except ValueError as e:
        assert str(e) == "Received only system message(s). "

async def test_agenerate():
    try:
        _ = await LLM._agenerate([prompt])
    except ValueError as e:
        assert str(e) == "Received only system message(s). "

async def test_astream():
    try:
        stream = LLM._astream([prompt])
        async for _ in stream:
            pass
    except ValueError as e:
        assert str(e) == "Received only system message(s). "
        
if __name__ == "__main__":
    test_generate()
    test_stream()
    asyncio.run(test_agenerate())
    asyncio.run(test_astream())

throws the errors, wasn't sure if you wanted tests but this works now