Retrieve Total Tokens by Trace ID using REST API or SDKs

Shashi D. · 2024-12-16T21:34:36.146Z

Hi Team, is it possible to get the Total Tokens based on a Trace ID using REST API or any of the SDKs? Any help is appreciated. Thanks.

Retrieve Total Tokens by Trace ID using REST API or SDKs | Arize AI Community

7 comments

· Sorted by Oldest

RunLLM
APP
·
It seems likely that retrieving the total token count based on a Trace ID might not be directly supported via a REST API or SDK in Phoenix. However, you can query token counts using the GraphQL API. The Project class in the Phoenix API provides fields like token_count_total, token_count_prompt, and token_count_completion which can be queried for a specific time range or filter condition, but it doesn't explicitly mention querying by Trace ID (source).Would you like more detail on how to use the GraphQL API to potentially achieve this? Sources: - Project.py
💡 Hint: Mention RunLLM in the thread for followups.
Dustin N.
·
Hi Shashi D. I'm sorry but our REST API and SDK do not currently support that feature, as for now we're only aggregating tokens dynamically on rendering full traces
Shashi D.
·
Thanks Dustin N.. Is the previous RunLLM response correct that it would be possible using the GraphQL API?
Dustin N.
·
yes, it's possible since that's how we render it, but our GraphQL API is primarily for internal use and I wouldn't rely on it being stable
🙏1
Dustin N.
·
hi Shashi D. just checking in—is this an approach you'd be interested in trying?

Shashi D.

Hi Dustin N., thanks for checking in. Yeah, I was able to query on the GraphQL playground with the following query to get the required details based on the TraceID. I was also able to confirm the same using the API key:

query MyQuery {
  projects {
    edges {
      node {
        trace(traceId: "4117e78a41be37dadfafbb6359036123") {
          traceId
          projectId
          spans {
            edges {
              node {
                tokenCountTotal
                latencyMs
                endTime
                startTime
                statusCode
                statusMessage
                tokenCountPrompt
                tokenCountCompletion
              }
            }
          }
        }
      }
    }
  }
}

Thanks a lot for the help. Really appreciate it.

Dustin N.
·
great! I'm glad

RunLLM
APP
·
It seems likely that retrieving the total token count based on a Trace ID might not be directly supported via a REST API or SDK in Phoenix. However, you can query token counts using the GraphQL API. The Project class in the Phoenix API provides fields like token_count_total, token_count_prompt, and token_count_completion which can be queried for a specific time range or filter condition, but it doesn't explicitly mention querying by Trace ID (source).Would you like more detail on how to use the GraphQL API to potentially achieve this? Sources: - Project.py
💡 Hint: Mention RunLLM in the thread for followups.
Dustin N.
·
Hi Shashi D. I'm sorry but our REST API and SDK do not currently support that feature, as for now we're only aggregating tokens dynamically on rendering full traces
Shashi D.
·
Thanks Dustin N.. Is the previous RunLLM response correct that it would be possible using the GraphQL API?
Dustin N.
·
yes, it's possible since that's how we render it, but our GraphQL API is primarily for internal use and I wouldn't rely on it being stable
🙏1
Dustin N.
·
hi Shashi D. just checking in—is this an approach you'd be interested in trying?

Shashi D.

query MyQuery {
  projects {
    edges {
      node {
        trace(traceId: "4117e78a41be37dadfafbb6359036123") {
          traceId
          projectId
          spans {
            edges {
              node {
                tokenCountTotal
                latencyMs
                endTime
                startTime
                statusCode
                statusMessage
                tokenCountPrompt
                tokenCountCompletion
              }
            }
          }
        }
      }
    }
  }
}

Thanks a lot for the help. Really appreciate it.

Dustin N.
·
great! I'm glad