ai-safety-foundation / sparse_autoencoder

Sparse Autoencoder for Mechanistic Interpretability
https://ai-safety-foundation.github.io/sparse_autoencoder/
MIT License
171 stars 39 forks source link

Add DDP support to metrics #196

Closed alan-cooney closed 7 months ago

alan-cooney commented 7 months ago

Switch all metrics and loss to extend torchmetrics. This will allow them to work across processes with DDP.