Arize AI Community Icon

Home
Events

Arize AX Releases
Arize AX Support
Arize News
Arize Observe
Discussions
Introductions
Phoenix Support

Powered by Tightknit

Arize AX Support

Arize AX Support

Reducing Latency in LLM Calls and Project UI Traces

·Jul 07, 2025 09:45 PM

What are some things I can do to reduce the latency between when I call an LLM in my application and when the trace appears in the project UI?

13 comments

· Sorted by Oldest

·

🔒[private user] Here is how I currently set up my instrumentation:

from openinference.instrumentation.anthropic import AnthropicInstrumentor 
tracer_provider = register(space_id=os.getenv("ARIZE_SPACE_ID"),
                 api_key=os.getenv("ARIZE_API_KEY"),
                 project_name=project_name)
        AnthropicInstrumentor().instrument(tracer_provider=tracer_provider)

Is this inline with documentation recommendations?

Brandon
·
Ya... I just don't understand why it's so slow
Xander S.
·
Hey Brandon, how long does it take?
Brandon
·
At least 3mins
👍1
Brandon
·
Maybe I'm abusing the feature?
Xander S.
·
It looks like you are using Phoenix Cloud?
Brandon
·
No, I'm using the Arize Cloud Platform. Do y'all refer to that as Phoenix Cloud?

·

FYI, I tried to disable the batch option in the call to register but that did not help

    tracer_provider = register(
        space_id=ARIZE_SPACE_ID,
        api_key=ARIZE_API_KEY,
        project_name=project_name,
        batch=False, # the default is true according to the source code
    )

Xander S.
·
No, I'm using the Arize Cloud Platform. Do y'all refer to that as Phoenix Cloud?
These are separate platforms.
Xander S.
·
I am not sure of what the standard ingestion times might be on the Arize Cloud platform (I am a maintainer on Phoenix).
Kinan F.
·
Can we get an answer on the normal ingestion times in the Arize platform and whether there are other options for using register that may expedite the export, processing and ingestion of the traces?
Kinan F.
·
Thanks Harrison. I just DMed you the project name
Brandon
·
Hello! I wanted to check back in on the expected behavior here. 🔒[private user]