Modelpedia
Work in progress
About
Findings
Models
Concepts
Methods
Datasets
Sources
Light
Dark
Residual Stream Analysis with Multi-Layer SAEs
2025-01-22
· ICLR 2025 Poster ·
anchor
Findings
IC-296
The degree to which SAE features are active at multiple residual-stream layers increases with model size in Pythia, Gemma 2, Llama 3.2, and GPT-2
IC-297
Applying tuned-lens transformations to the residual stream decreases the apparent multi-layer SAE feature activity from 54–88% to 37–41% of total variance