Export
Trainer callback that periodically writes model checkpoints to a
models directory under the run’s output directory, so training can be resumed and
trained models can be used for inference.
Configuration
Option |
Description |
Type |
Default |
|---|---|---|---|
|
Number of epochs between checkpoint writes. |
int |
|
trainer.register_callback(CallbackSpec(callback=ExportCallback, kwargs={"save_interval": 10}))