Description
Understanding how an LLM thinks internally, like what one attention head does, is hard to dig into with ordinary frameworks. TransformerLens is a library for mechanistic interpretability of GPT-style language models.
It loads dozens of open models and makes reading and editing activations at any layer easy, a standard interpretability tool.
Activations:Any layer.
Hooks:Intervene mid-run.
Models:Dozens supported.
Standard:Community tool.
It loads dozens of open models and makes reading and editing activations at any layer easy, a standard interpretability tool.
Features
Activations:Any layer.
Hooks:Intervene mid-run.
Models:Dozens supported.
Standard:Community tool.
