Getting Started with fastxtend Callbacks
As an extensions library for fastai, fastxend contains multiple types of callbacks. This guide will briefly go over all types of callbacks in fastxtend. Every callback in this guide will link to the full callback documentation.
- New and replacement callbacks
- Task specific callbacks
- Modification callbacks
- Utility callbacks
- Feature porting callbacks
New and Replacement Callbacks
The first type of callback in fastxtend are new and replacement callbacks. These callbacks either add new features to fastai or augment existing fastai functionality but are fully backwards compatible. These callbacks are imported via task specific all imports such as fastxtend.vision.all or fastxtend.text.all.
New Callbacks:
- Exponential Moving Average Callbacks
- Tracking Callbacks
Replacement Callbacks:
- CutMixUp Callbacks
- Learning Rate Finder Callback
Task Specific Callbacks
These callbacks are only imported with task specific imports.
Vision callbacks imported from fastxtend.vision.all:
- CutMixUp Callbacks
- Progressive Resizing Callback
Text callbacks imported from fastxend.text.all:
Audio callbacks imported from fastxtend.auido.all:
Modification Callbacks
These callbacks modify either existing fastai functionality or change the fastai defaults and must be imported separately.
- PyTorch Compile Callbacks imported via
fastxtend.callback.compiler - Gradient Accumulation Callbacks imported via
fastxtend.callback.gradaccum - Profiling Callbacks imported via
fastxtend.callback.profiler
Utility Callbacks
These callbacks add tools for developing other callbacks. - LogDispatch - CallbackScheduler
Feature Porting Callbacks
These callbacks backport current fastai features to older versions of fastai. These callbacks started out as fastxtend callbacks and then were upstreamed into fastai. If you are using the latest version of fastai you will be using the fastai versions of these callbacks.