As AI packages grow to be extra advanced, are other folks in fact monitoring token utilization and prices at a workflow degree?
It is simple sufficient to look utilization for person fashion calls, however as soon as a function spans a couple of activates, fashions, gear, retries, and background jobs, I have discovered it a lot tougher to respond to questions like:
-
Which workflow is using prices?
-
The place is latency being offered?
-
Which step failed?
-
How a lot does a unmarried consumer motion in fact value?
Curious what others are the use of as of late.
