Sweeps over hyperparameters, once subnetwork probing metrics are up.

DavidUdell / sparse_circuit_discovery

Circuit discovery in GPT-2 small, using sparse autoencoding

MIT License

7 stars 1 forks source link

Sweeps over hyperparameters, once subnetwork probing metrics are up. #17

Closed DavidUdell closed 10 months ago

DavidUdell commented 10 months ago

Probably with wandb logging.

DavidUdell commented 10 months ago

As with sparse autoencoding itself, it's own repo for these sweeps. Sweeps in this repo will have to be qualitative, or otherwise use the metrics that fall right out of systematic ablations in a basis.