neuron-explorer: open-source steering-first neuron explorer for Qwen3.5-4B
viksit · x · 2026-10-26
An open-source (531 stars) steering-first neuron explorer for Qwen3.5-4B: it pushes every MLP neuron in a layer up and down on a few prompts, keeps every rollout, and lays results over a 3D view of the model.
It loads the model locally (MPS/CUDA/CPU) for live steered chat, lens, connections and live inference. Setup: repo ships with caches; run uv sync then uv run python explorer/server.py (localhost:8791), with sweep scripts producing runs/ and stats/ data. A practical starting point for mechanistic interpretability and activation steering experiments.
More from Research
- Zeroing 16 neurons makes a typical LLM reply 830,000× less likely — Sauers_ · 2026-10-26
- FOCS '26 paper closes the gap on information complexity of VC classes — roydanroy · 2026-10-26
- Steering a single neuron expands a model's J-space via J-lens-shaped residual stream — Sauers_ · 2026-10-26
- UniForge3D turns sparse-view inputs into detailed 3D meshes, code released — mervenoyann · 2026-10-26
- 30% of US STEM PhD dissertations now contain AI-generated text, NBER paper finds — 0xkarasy · 2026-10-26
- None of the 11 grand challenges in single-cell science solved since 2020 — Sauers_ · 2026-10-26