TransformerLens
What it is
A library designed for the mechanistic interpretability of GPT-style language models. It allows users to load over 15,000 open-source models and inspect internal activations, providing tools to edit, remove, or replace those activations during model execution.
The problem it solves
Provides a unified interface to access and manipulate internal activations across a wide variety of model architectures and families.
What you could use it for
- Reverse engineering algorithms learned by LLMs from their weights
- Analyzing induction heads and in-context learning behaviors
- Probing and identifying specific neuron representations in large language models
- Automated circuit discovery for model interpretability
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.