lyramakesmusic / neuron-explorer

Why you might care

For interpretability researchers who want to probe a single 4B model neuron by neuron: it runs steered rollouts at several doses, stores them, and offers lens, connection and live steered chat views. It needs a local Qwen3.5-4B and a CUDA GPU pod with a vLLM fork for the sweep, and the caches ship with the repo so the explorer can run without the sweep.