Monitor PyTorch model internals

To monitor PyTorch model internals, use Neptune TorchWatcher.

It's a lightweight monitoring tool that automatically tracks layer activations, gradients, and parameters during training and logs them to Neptune. It allows you to define the type of metadata to track and organize it within the hierarchical namespace structure.

Example use cases:

• Track all metadata during initial training.
• Track activations during validation phase.
• Track gradients during later stages of training.

The README file contains prerequisites and complete instructions.