Improve autoencoder training
Move constants and SparseAutoencoder class to own file
Move hyperparameters to own file
Update .gitignore

Add visualization of sae feature activations for a few sample data points. Does not work properly yet

Add DataLoader class to load activations

Improve calculating activations
Scale down model and therefore compute and disk requirements
Reduce reserved memory for slurm task, as it was not needed

Add calculate_activations.py
Update .gitignore
Implement retrieving activations from model for data

Add .gitignore
Add run shell files for slurm
Implement loading and querying Phi2
Implement downloading and preprocessing data