Libris Translate 0.28.0 · User guide
Category: Manage Libris Translate
Usage and request history
See what the models did, what it cost, and how far to trust the estimates.
Steps
- Open the book’s Observability tab. “Model requests” lists every call: time, operation, model, status, duration, input and output tokens, throughput, attempt and cache. Click one for its “Request details”.
- While a job runs, “Active work estimates” gives the estimated remaining time and cost, with an “Estimate confidence” that grows as successful requests are measured.
- “Wasted input tokens” shows the share of input lost to invalid responses or errors, which often explains why the real cost exceeds the first estimate.
- Administrators open Statistics in the sidebar for usage “By model” across the installation and “Failures by cause”, grouped by provider.
What to know
Costs are list-price costs computed from the prices you entered, not the provider’s invoice. Tokens a provider does not report are recorded as zero. By default, the text of prompts and answers is kept for 30 days, while the usage figures remain. Statistics and Settings are reserved to administrators.
Fictional example
In a fictional book, compare a successful request with a retried one to see where the extra input tokens went.