calibrate¶
Run the model calibration benchmark. Foil uses the result to pick a good concurrency and batch size for the current model + hardware combination.
foil calibrate¶
Run model calibration benchmark.
Usage:
Options:
When to run it¶
- After downloading a new model and activating it
- After upgrading Foil or macOS
- If scans feel unusually slow (run with
--showfirst to see the last calibration result)
Calibration takes a few minutes and writes its output to ~/.foil/calibration.json.