Nouamane B.

Ok great, that clears it up so that isn't a huge change for us luckily Yes we could, but what we really want to achieve is having it running to always collect traces within our AWS infra However, for local testing absolutely that's what we'll do, and point it to a separate table in that RDS instance 🙂 Thanks for you help!

We want to use that tunneled connection I mentioned earlier and connect that to a local instance of the UI. This way we don't need to worry about managing a UI interface, so when we want to check out what the traces look like we run a local launch of the UI that connects to the real database

I see, so the thing I noticed is that the phoenix container ships with both UI and server right? Is there a container for deploying just the server?

Whoever setup the RunLLM app - great idea 😄

Works 🙂 Thanks for your help

I will try this thanks Dustin N.

Ah I understand

Or would I have to build it out custom?

We have a generator class which is just abstraction of a small chat pipeline

from openinference.instrumentation.haystack import HaystackInstrumentor
from openinference.instrumentation import using_attributes

temperature = 0.9

print(f'registering tracing with initial temprature of {temperature}')
tracer_provider = register(
  project_name="default",
  endpoint="https://app.phoenix.arize.com/v1/traces"
)

HaystackInstrumentor().instrument(
  tracer_provider=tracer_provider
  )

def main():
    print('setting up generator')
    prompt = 'What is the capital of Peru'
    generator = GeneratorService()
    # generator.generate(prompt=prompt)
    print('sending to generate with pipeline')
    print(generator.generate_with_pipeline(prompt=prompt, temperature=temperature))

main()