Pinned Loading

1. activation-patching-framework activation-patching-framework Public

   Causal intervention framework for mechanistic interpretability research. Implements activation patching methodology for identifying causally important components in transformer language models.

   Python 1

2. logit-lens-explorer logit-lens-explorer Public

   Mechanistic interpretability tool visualizing GPT-2's layer-by-layer predictions using the logit lens technique

   Python 2

3. induction-head-detector induction-head-detector Public

   Mechanistic interpretability tool to detect induction heads in GPT-2 using TransformerLens

   Python 1

4. minigpt-shakespeare minigpt-shakespeare Public

   From-scratch GPT implementation trained on Shakespeare (10.75M params, 25 tests)

   Python 1

5. bpe-tokenizer-scratch bpe-tokenizer-scratch Public

   Byte-Pair Encoding tokenizer built from scratch in Python. The same algorithm used by GPT-2.

   Python 1

6. toy-model-superposition toy-model-superposition Public

   A minimal Python implementation replicating Anthropic’s “Toy Models of Superposition,” illustrating feature superposition under sparse activations.

   Python 1