SONAR SAEs
Collection
Sparse Auto-Encoders for SONAR sentence embeddings, from Pochinkov & Darmawan (2025) (EACL submission). • 5 items • Updated
SONAR SAEs — wandb logs
Raw wandb run directories (config, summary, system metrics, scalar training curves) matching the SAE checkpoints in:
nickypro/sonar-saes-large— scaled-up BatchTopKnickypro/sonar-saes-comparison— four-variant comparison
Run IDs in directory names match the checkpoint run IDs in those repos.
Companion paper
Interpretability of Text Auto-Encoders using Sparse Auto-Encoders: A Sandbox for Interpreting Neuralese. Nicky Pochinkov & Jason Rich Darmawan, EACL 2026 (submitted).
Use
The original wandb runs were logged to the
seperability wandb team. This repo
mirrors the on-disk run directories for archival reproducibility; for
live dashboards, see the wandb project directly.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support