Open nckxneoro52449 opened 1 month ago
At present we do not include this functionality, but there are active plans to do so in future: See e.g. https://github.com/Cambridge-ICCS/FTorch/discussions/22
It is something we hope to look at soon after we finalise GPU functionalities make a first release, but community contributions are also welcome. Even if it is just a list of the functionalities you would like to see made available through FTorch.
To follow up on this, can you clarify how exactly you plan to use this?
We have an option requires_grad_opt
in the tensor and model loaders that turn on gradient calculation, and is_training_opt
for models.
See: https://cambridge-iccs.github.io/FTorch/interface/torch_tensor_from_array.html https://cambridge-iccs.github.io/FTorch/proc/torch_module_load.html
If either of those help.
I have a question regarding the functionality of Ftorch. Does Ftorch currently support PyTorch's autograd feature for calculating first-order derivatives (Jacobians)? If not, are there any plans to add this feature in the future?