#llm-internals

Qwen-Scope

Sparse autoencoders that turn LLM black-box internals into interpretable features you can actually use.