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
liked a dataset about 15 hours ago
msakarvadia/handwritten_multihop_reasoning_data liked a dataset about 15 hours ago
rmovva/HypotheSAEs liked a model about 15 hours ago
RobinHaselhorst/AMF-harrypotter-7b