Explanation of the InputTransform

❓ Questions and Help

Is there an explanation of the InputTransform and how to use it anywhere?

I see that there are some basic examples under the documentation of some of the tasks. However, what I am missing is:

1. An explanation of what the *_on_device() functions refer to in contrast to the standard functions
2. How the various functions are combined? Many available functions overlap such as per_sample_transform(), input_per_sample_transform() and train_input_per_sample_transform(). Is simply the more specific function used or will the per_sample_transform() always be applied first before applying the more specific functions?

[tchaton]

By default, the data in Flash are dictionarry. By using input_*, this would apply your function on the input key. And *_on_device is applied when the data are already on the gpus, tpu, etc… And yes, always the most specific method is used

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.