Tutorials

Tutorials cover traceTorch mechanics that are useful after you have built a first model.

Read these when you need to preserve hidden state, optimize a trained model for inference, or create your own traceTorch-compatible layer.

Contents:

• Saving and Loading States
  • Why states are separate
  • Saving
  • Loading
  • Shape checking
  • Continuing a sequence
• Compiling and Decompiling
  • Basic use
  • Returning to training
  • What gets compiled
  • State and compilation are separate
  • Checking equivalence
• Creating a Custom Layer
  • The goal
  • Subclass a traceTorch layer
  • Registering a state
  • Forward pass
  • Using the layer
  • Adding constrained parameters
  • Checklist