Easily evaluate models steered by SAEs#2641

Open

AMindToThink wants to merge 8 commits intoEleutherAI:mainfrom AMindToThink:sae_steered

+426

Commits on Jan 20, 2025

Can evaluate sae_steered_beta models
AMindToThink
committed
Can evaluate sae_steered_beta models add
AMindToThink
committed
Ran pre-comit and pytest
AMindToThink
committed
Add larger description for feature in dog_steer.csv
AMindToThink
committed

Commits on Jan 21, 2025

Cache the repeated SAE instead of making new every time.
AMindToThink
committed

Commits on Jan 22, 2025

Types of hook in InterventionModel
AMindToThink
committed
Other steer functions. Deprecated InterventionModel instantiation from hooks list; now only csv
AMindToThink
committed

Commits on Jan 23, 2025

Other hook functions
AMindToThink
committed