MechInterp Decoder Analysis
· Batch Import
Description
Analyze SAE (Sparse Autoencoder) decoder weights to understand feature output influence, importance, and similarity. Reveals what outputs a feature promotes or suppresses, providing complementary insights to activation analysis for interpretable AI.
Repository
https://github.com/majiayu000/claude-skill-registry/tree/main/skills/development/mechinterp-decoder
View on GitHub