Logging, Traces and Sharing Output

I've been working with prospects in building out simple RAG setups and realized that what I was presenting didn't have adequate logging.

Logging in AI isn't just for validation of the outputs, but it also functions as future training.

I typically use Llama Index as my go to framework for building rag pipelines. I've tried WandB previously for logging, it didn't seem as simple to me. So, i tried out Langsmith. Langsmith seems like a simpler tracing and evaluation framework to work with. If you're trying to get started, it's very simple, just use their python SDK if you're not a native Langchain user, https://docs.smith.langchain.com/tracing/quick_start .

What i like the most thus far is that I can easily share particular runs from my Langsmith dashboard, have a look -> https://smith.langchain.com/public/acde7dac-5b11-47d2-a9d6-43faa805b7cc/r

What do you think about logging and tracing? Do you have a favorite framework?

Let me know in the comments!

0 comments