search

DavidUdell / sparse_circuit_discovery

Circuit discovery in GPT-2 small, using sparse autoencoding
MIT License
7 stars 1 forks source link

Batching in `feature_web_webtext`. #19

Closed DavidUdell closed 10 months ago

DavidUdell commented 10 months ago

The shape of the activations tensor cached by hooks_manager will change, but this should be possible, because the select subset of the dataset is all facing a single ablation hook at one time. This is a runtime bottleneck.

DavidUdell commented 10 months ago

Actually, upside here is more limited than I had hoped.