Srilakshmi C.

Hi Nas, I think the bot here is mostly correct! To answer your questions, 1) message.reasoning_ is unfortunately not supported today in open inference but you can make a ticket here! 2) From my understanding, I think you can do the 2 message workaround. I think another potential could be making a child span with the reasoning content? meaning, a child span named reasoning with its own llm.output_messages[0].message.content 3) I would suggest opening an issue abt this into the openinference repo as an enhancement! Hope this helps & let me know if there's anything else we can help with!

Hi Adam H. , Prompt name will map to one prompt ID. within a prompt, if you have multiple prompt versions, each of those can be differentiated with the version_id. for how to get these prompts into your code, you can run this code block, but replace with your prompt name:

from arize.experimental.prompt_hub import ArizePromptClient, Prompt, LLMProvider

client = ArizePromptClient(api_key=os.environ["ARIZE_API_KEY"], space_id=os.environ["ARIZE_SPACE_ID"])

//replace the line below with your prompt name
prompt = client.pull_prompt("test-basic")

//or if you want to pull all your prompts, you can run the following line
prompts = client.pull_prompts()

more info on the code functions here: https://arize.com/docs/ax/reference/reference/prompt-hub-api#core-components In the UI, it might look something like this. The prompt ID & prompt name, found in the top left, will stay the same regardless of which version you are looking at. The version ID will change from version to version.

Hi Shubham A., you can get your annotations by pulling down your trace dataset in AX & your annotations should be in that dataset by default. Here's a code snippet that might help:

from arize.exporter import ArizeExportClient
from arize.utils.types import Environments
from datetime import datetime, timedelta, timezone

client = ArizeExportClient(api_key=os.environ["ARIZE_API_KEY"])
model_id = "[insert project name]"

primary_df = client.export_model_to_df(
    space_id=os.environ["ARIZE_SPACE_ID"],
    model_id=model_id,
    environment=Environments.TRACING,
    start_time=datetime.now(timezone.utc) - timedelta(days=365),
    end_time=datetime.now(timezone.utc)
)

primary_df['annotation.test.label']

Hope this helps for now! Please feel free to ask any more questions you have

Hi Chris, the bot here is correct -- in terms of just setting up your phoenix instance, there's a number of ways to do so here. & then from there you'll want to manually instrument your application - the best resources I can recommend for getting started there is this external example set & our documentation for how to setup tracing. Hope this helps a little bit and points you in the right direction! Feel free to ask any other question or share any issues you run into!

hey Ben, could you maybe make a ticket here for this enhancement, just so we can track it and use it as reference moving forward: https://github.com/Arize-ai/phoenix/issues ?

Hi @/Matthew Ropp, I also don't have experience with setting up airflow tests but happy to take a look. Would it be helpful to grab some time to chat through your use case and how I can help best!

Hi Akshay P.! Thank you for pointing that out, we will try to get that up and running shortly. cc Sanjana Y.

Hi Keerthi! Currently in the UI we have a way to filter spans from a certain time frame and then export those from a project into a dataset. (I have included a quick walkthrough in the GIF below)

Hi Peter P. ! Thank you so much for your question -- I can replicating the issue and seeing how to solve it. this code that you have here:

client = openai.OpenAI()
    response = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{"role": "user", "content": "Say 'Hello from minimal test'"}],
        max_tokens=10
    )

won't be picked up by the instrumentation here since it is designed to patch OpenAIAgents:

instrumentor = OpenAIAgentsInstrumentor()
instrumentor.instrument(tracer_provider=tracer_provider)

instead, please try using:

from openinference.instrumentation.openai import OpenAIInstrumentor
OpenAIInstrumentor().instrument(tracer_provider=tracer_provider)

Please let me know how this works for you!