Tag
The author built a custom llama.cpp server and Mikupad UI to enable local inference and activation steering with Anthropic's open-weight Natural Language Autoencoders. A LoRA version is in development to reduce memory requirements.