[HuggingFace] ELIZA with Sparse Autoencoder

Part of

Inaudible Speech LLM Vulnerabilities