Skip to main content

Ctrl+K

Sparse autoencoders

Sparse autoencoders from scratch
SAE architectures & JumpReLU — the activation GemmaScope uses
Reconstructing a Neuronpedia dashboard, panel by panel
Training your own JumpReLU SAE on Gemma 3 1B

Training data attribution

Influence Functions: Where Does the Hessian Come From?

ML

Forward, backward, and hooks in PyTorch

Appendix

References

Index

By J Rosser

© Copyright 2026.