Transformer Debugger
Component can be an attention head or neuron, or autoencoder latent
Each node only exists in one forward/backward pass. If you modify the prompt and rerun, that would create different nodes
Residual stream stands for
Circuit is a set of nodes that work together to perform some behavior/reasoning.
TDB Usages
git clone https://github.com/openai/transformer-debugger tdb cd tdb python -m venv ~/.virtualenvs/transformer-debugger source ~/.virtualenvs/transformer-debugger/bin/activate pip install -e . pre-commit install cd neuron_viewer
