Anthony P.

I am glad to hear it, and would be very curious to know the root cause

Thanks, I will look into any potential incompatibilities there. That operating system is not within our testing matrix

Do you know which operating system + version your phoenix instance is running on?

thank you 🙏

Can you share the command you use to launch the phoenix server in your instance?

I have created various testing scripts using litellm + bedrock and the same model name that you are using, and I cannot replicate the issue; I see cost details for each of my spans. I will have a short period of availability from 2:00pm est that we can try for a call. In the meantime, if you can run this test script and verify that the spans produced also do not display cost in Phoenix that would be very helpful. You may have to tweak the Phoenix connection arguments in the instrumentation if you are not running a local phoenix instance or have authentication enabled. The script uses uv, but you may be able to execute with python directly if you have the required dependencies already installed in a virtual env. uv run litellm_bedrock_anthropic_async.py

In the meantime, could you run the following graphql query in the "api/graphql" section of the UI and paste the results back here?

{
  generativeModels(first: 100) {
    edges {
      node {
        id
        name
        namePattern
        provider
        startTime
        kind
        tokenPrices {
          tokenType
          kind
          costPerMillionTokens
          costPerToken
        }
      }
    }
  }
}

I will get back to you on a time that works for me today, let me make sure I have a clear picture of how cost processing works end to end so that I can make effective use of our time on a call

In my testing the model name visible in the span you sent does work with the default regex pattern. Are you running phoenix via the python command or docker out of curiosity? And what python version are you using? I am not available for a call today unfortunately

did you happen to try submitting spans without making a custom model definition first? it worked out of the box for me without a custom model regex