Factorizing a model's final unembedding into a sparse feature basis for logit-lens analysis.
🤝 Open to Collab
Matteo He
hematteo
AI & ML interests
Generative AI, Mechanistic Interpretability, AI Safety, Post-Training, Reinforcement Learning
Recent Activity
upvoted a paper 3 days ago
Learning to Read Out: Unembedding Dynamics in Language Model Pretraining liked a dataset about 1 month ago
msakarvadia/handwritten_multihop_reasoning_data liked a dataset about 1 month ago
rmovva/HypotheSAEs