Open dkgaraujo opened 2 years ago
With the release of keras_core
and its subsequent move into the main keras
library, I will experiment with first passing PyTorch-based models into a keras
model and then using that to bind with gingado
. This is made possible now that keras
models also accept torch
layers. In addition to simpler maintenance, it would offer the user the possibility of changing the backend more easily and of loading the data more flexibly (using PyTorch dataloaders
or the respective TensorFlow object).
The base class
ggdModelDocumentation
can read characteristics from gingado, scikit-learn and keras models. But PyTorch and fast.ai models are not yet supported. Ideally this should be included.