Assistant tokens information incorrect

kslays

In the Assistant, clicking the circled i shows more detail about the query. I noticed the tokens, total time, and tokens per second don't correspond. Why not?

The tokens per second should match the tokens/time, or if not, an explanation might be useful. I could understand the tokens per second being a higher number than tokens/time (if there were startup and other latencies) but it doesn't make sense to me for it to be slower.

yiwei-1

Hi, to clarify, the token count you’re seeing includes both input tokens (query, system prompt, tool calls, etc.) and output tokens, while the speed measures only output tokens. Perhaps it would be better to show a more detailed breakdown of input and output tokens here?

kslays

Yes, listing input and output tokens separately would solve this. Honestly, I appreciate detail so accounting tokens to each source would be ideal for me (user, each tool, thinking, and output). If that were implemented, then a total would also be helpful.