[Tooling] Show input/output token count in Dev UI
Describe the bug Input/output token counts are tracked but are only visible in GCP. It would be nice to see the input/output counts directly in the trace in Dev UI to get an idea for how expensive the call is during development.
Edit: Token counts are available in the output section of the trace viewer (e.g. on the model action).
- For the trace view, we should consider add badging and/or other UI affordances to elevate this information into clear sight.
- We could also consider displaying token counts on the model and prompt playgrounds as well.
Typically I am able to see these values in my local traces. Can you clarify?
Ah, I didn't realize. I kind of expect it to appear everywhere latency does (left sidebar, right sidebar), maybe with parent nodes summing child nodes so that you can get a feel for costs at a glance just like you can with latency. WDYT?
Definitely - that sounds good to me (badges for tokens, etc). Thanks for clarifying.