po13on

No description.

A Python research workspace for probing how Qwen language models handle Theory of Mind reasoning, using TransformerLens to capture internal activations and trace which attention heads contribute most to nested belief-state answers.

No description.

[NeurIPS 2024 Spotlight] Code and data for the paper "Finding Transformer Circuits with Edge Pruning".

Code for "Automatic Circuit Finding and Faithfulness"

A library for mechanistic interpretability of GPT-style language models

Accessible large language models via k-bit quantization for PyTorch.

Config files for my GitHub profile.