Added support to get usage for ChatGoogleGenerativeAI

[NeerajG03]

This PR has been raised to add support to track ChatGoogleGenerativeAI's langchain call usage.
A simple check for usage_metadata attribute and then parsing using the _parse_usage_model

[CLAassistant]

All committers have signed the CLA.

[greptile-apps]

Disclaimer: Experimental PR review

PR Summary

Added support for tracking usage of ChatGoogleGenerativeAI's langchain calls in the Langfuse Python SDK.

• Introduced check for 'usage_metadata' attribute in message chunks in langfuse/callback/langchain.py
• Implemented parsing of 'usage_metadata' using _parse_usage_model function
• Enhanced usage tracking capabilities for Google's Generative AI in addition to existing LLM providers

_{1 file(s) reviewed, no comment(s)
Edit PR Review Bot Settings}

[maxdeichmann]

@NeerajG03 did you test this locally? Thanks a lot for the contribution!

[NeerajG03]

@maxdeichmann I'm sorry i'm very new to contributing for projects.
If you are asking if i have run this test poetry run pytest -s -v --log-cli-level=INFO, then thats a yes. 👍
I ran it for test_langchain.py and test_langchain_integration.py

[NeerajG03]

Here is the screenshot of the run summary

[aunitt]

This also worked for me for ChatVertexAI.

It appears to fix my issues which appear to be those mentioned in langfuse/langfuse#5468

[ReneFabricius]

I would suggest adding

Suggested change

	break
	if llm_usage:
	break

or even better modify the previous condition to:

                if generation_chunk.generation_info and (
                        generation_chunk.generation_info.get("usage_metadata", None)
                    ):

as in my testing with ChatVertexAI generation_info contains empty usage_metadata while message contains valid one.

[ReneFabricius]

Looking at it better, this should have no effect, as a few lines down, there is:

or getattr(message_chunk, "usage_metadata", None) # for Ollama