Cached prompt tokens
Prompt tokens served from cache in the day, as the platform counts them (Surplus
cache_tokens). Part of the input in inference_tokens. Daily, UTC; the running day is partial.Definition
At a glance
- Key
inference_cached_tokens- Unit
- Tokens
- Aggregation
- SumA flow: weekly and monthly values add up the daily values in the period.
- Can be negative
- NoNegative values are treated as data errors and held back by the range check.
Sources
Where it comes from
Datasets that write this metric. When several sources provide it, each chart states which one it uses.
| Dataset | Source |
|---|---|
Surplus tokens, cache, requests and savings (daily)surplus.usage | Surplus Intelligence |
Usage
Dashboards using it
No published dashboard uses this metric yet.
Time buckets, partial periods and restatements follow the site-wide conventions.