Restore `top_tokens.py`--it's too hard right now to guess what features do

DavidUdell / sparse_circuit_discovery

Circuit discovery in GPT-2 small, using sparse autoencoding

MIT License

7 stars 1 forks source link

Restore `top_tokens.py`--it's too hard right now to guess what features do #57

Closed DavidUdell closed 8 months ago

DavidUdell commented 8 months ago

This would save to a different script than TOP_K_INFO_FILE.YAML. But it would be human interpretable, unlike that file in the current interp pipeline.

DavidUdell commented 8 months ago

Neuronpedia (https://www.neuronpedia.org/gpt2-small) has me covered!