Hello, By any chance, does anybody have made a docker image that launches phoenix with a volume to persist the traces?
Hey Didier, thanks for checking out Phoenix! Phoenix is currently a notebook application and we don't have a Docker image, although we are exploring ways to persist trace data. Can you tell us more about how you're using Phoenix and what your ideal flow would look like?
Hi Xander! I'm using Phoenix as an alternative to W&B for visualizing experiments I do with LangChain. While W&B is very complete, I find there UI for LLM traces is not intuitive to navigate, and misses information like token usage (or at least, it is not easy to find). So far I love what I see with Phoenix, thank you for all the work :) Ideally, I would like a pre-built Phoenix docker image, which I could easily connect to my application's docker via compose, and that would record all the traces. I have not explored that part much yet, but being able to label the traces with a dict that contains configuration information from my application, or anything that would help filter out the traces later on. I understand that I can have different sessions, but not sure about the rest. Also, a simple CLI would be nice, something with commands like phoenix server start to run the tracing server or phoenix view <dataset-name> to launch the server for a given dataset name.
Hey Didier M. - we'll get you a docker file soon. As Xander mentioned, we started with the in-notebook experience but have some hosting / persistence work coming up. Will try to keep you informed. Also feel free to drop us enhancement requests! We'd love to hear from you on GitHub and we'll notify you as we work on them. Thanks again for the feedback!
hey Didier M. just published one - let me know if you give it a try! https://hub.docker.com/r/arizephoenix/phoenix
