[P2] Add Sparse Autoencoder Interventions #164
Merged
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
Add an
AutoencoderLayer
and anAutoencoderIntervention
to support interpretability methods that use autoencoders to learn interpretable feature space, including Sparse Autoencoders.AutoencoderLayer
defines any autoencoder with a single-layer encoder and a single-layer decoder. Users can additionally define customized autoencoders by extending the base classAutoencoderLayerBase
.AutoencoderIntervention
defines an intervention that allows interchange interventions in the latent space of the autoencoder.The
AutoencoderIntervention
supports loading pre-trained autoencoders trained outsidepyvene
framework, with theget_intervenable_with_autoencoder
function below:The resulting intervenable, including the intervention dimensions and the autoencoder, can be saved as:
Fix #77
Testing Done
[internal only] https://colab.research.google.com/drive/1_fxM7JUqkMy6Erz6K1JV0NwQBw1r8g0k?usp=sharing
Will add this colab as a tutorial.
Checklist:
[Your Priority] Your Title