Tests still need to cover the Loss functions used in the training routine, this PR addresses this.
This PR also adds forces to the getter method of the TorchDataset and the collateconformers method, as raised here: #127 . It also improves performance by casting dtype on the full tensor and not on individual tensor slices.
Description
Tests still need to cover the Loss functions used in the training routine, this PR addresses this.
This PR also adds forces to the getter method of the
TorchDataset
and thecollateconformers
method, as raised here: #127 . It also improves performance by casting dtype on the full tensor and not on individual tensor slices.Todos
Status