ckkissane

attention-output-saes

Code to reproduce key results for "Interpreting Attention Layer Outputs with Sparse Autoencoders"

base-models-refuse

Code to reproduce key results accompanying "Base LLMs refuse too"

crosscoder-model-diff-replication

Open source replication of Anthropic's Crosscoders for Model Diffing

deep_learning_curriculum

Language model alignment-focused deep learning curriculum

rlhf-shakespeare

Shakespeare transformer fine-tuned to generate positive sentiment samples using RLHF

sae-dataset-dependence

sae-transfer

Code to reproduce key results accompanying "SAEs (usually) Transfer Between Base and Chat Models"

sae_vis

Create feature-centric and prompt-centric visualizations for sparse autoencoders (like those from Anthropic's published research).

shakespeare-transformer

Decoder only transformer trained on the works of Shakespeare

TransformerLens

ckkissane/numpy

The fundamental package for scientific computing with Python.