Interpreting Transformers thru Sparsity

Aiming to map neuron firing patterns of ReLU-based T5 models to human interpetable concepts.

Name		Name	Last commit message	Last commit date
Latest commit History 39 Commits
__pycache__		__pycache__
activating-prompts		activating-prompts
plots		plots
tokens-firing-neurons		tokens-firing-neurons
.DS_Store		.DS_Store
README.md		README.md
activating-prompts.pkl		activating-prompts.pkl
automated-interpretation.py		automated-interpretation.py
chatgptinterpretations.ipynb		chatgptinterpretations.ipynb
experiments.ipynb		experiments.ipynb
experiments_q.ipynb		experiments_q.ipynb
interpretable_neurons_100_200_d12.pkl		interpretable_neurons_100_200_d12.pkl
interpretable_neurons_100_200_e11.pkl		interpretable_neurons_100_200_e11.pkl
plotting.py		plotting.py
prompt_retrieval.py		prompt_retrieval.py
prompts.txt		prompts.txt
squad_layer_sparsity.pkl		squad_layer_sparsity.pkl
top-10-neurons.pkl		top-10-neurons.pkl
top_k.py		top_k.py

Provide feedback