Pytorch with Clara AutoML

I have a Python script training and evaluating my Pytorch model (custom model called DA-net) and I would like to use the AutoML functionality to search for my optimal hyperparameters. Is it possible to use Clara AutoML without the need for transferring my Pytorch model to Tensorflow (.graphdef, pb, … as used by NGC)? Or without the need of rewriting my model? I currently have .pth and .pt files.

Thanks

Hi
Unfortunately not in the current release. Clara train uses tensorflow as a back end and is built on NGC TF container. Therefore, all features for training as Federated learning and AutoML, multi-GPU training are only supported for tensorflow. In Future releases, we will be supporting PyTorch so stay tuned

Please note that AIAA supports both tensorflow and PyTorch models.

1 Like