Shunchang

/

sae-rm-checkpoints

like 0

Hi, I'm Niels from the community science team at Hugging Face. This pull request adds a model card for the Sparse Autoencoder (SAE) checkpoints associated with the paper "Preference Instability in Reward Models: Detection and Mitigation via Sparse Autoencoders".

The model card includes:

• Metadata for sae-lens and the feature-extraction pipeline.
• Links to the research paper and the official GitHub repository.
• A description of the reward models targeted by these SAEs.
• Sample usage for downloading the checkpoints using huggingface_hub.