Priyan J.

Hi Nouamane great question. If you are trying to optimize a multi-turn conversation, the judge LLM would definitely benefit from looking at conversation history. The best way to do this is to pass in traces into the meta prompt. That way, the meta prompt LLM (and judge LLM) have access to the entire path taken by the agent (tool calls, retrievals, all LLM calls) and can optimize the prompt with respect to that

Zhuyi X. which version of openinference-instrumentation do you have? my code actually still works with set_tracing_disabled(disabled=True). Sorry for the confusion.

Victor L. We have fixed this. You should be able to create an Arize service through Azure now.

Hi Victor L.. Our team is still working on some errors we see. I would suggest you create an Arize AX account for now and work through that. I will get back to you hopefully by end of week.

Hi stephen You can now use PHOENIX_LOGGING_LEVEL to toggle the Arize logger configuration. Released in arize-phoenix 10.7.0 You can set PHOENIX_LOGGING_LEVEL=WARNING to get rid of those INFO logs from the Uvicorn logger.

Yup, looking into it

My understanding is that PHOENIX_LOGGING_LEVEL works on Phoenix's own logs, but the /healthz messages come from a completely different logger (uvicorn.access) which Phoenix doesn't re-configure. but let me double check

default is still logs enabled

just adding an environment variable in case users do not want to see Uvicorn logs, as they can be expensive Stephen said Yeah, we are running Phoenix in prod so /healthz gets hit frequently. We can filter them out at query time, but would prefer to not be paying our logging provider for them

https://github.com/encode/uvicorn/blob/master/uvicorn/config.py, line 392 access_log toggles the logger named uvicorn.access that writes a one-line record for every HTTP request. it is probably also worth to add an environment variable for Uvicorn's log_level which could be set to warning for example.