neuron-explorer adds MLX backend for per-neuron steering on Apple Silicon
arthurcolle · x · 2026-10-27
arthurcolle added MLX support to neuron-explorer, an open-source steering-first interpretability tool for Qwen3.5-4B that pushes every MLP neuron in a layer up and down, keeps all rollouts, and browses results on a 3D model view. It runs natively on Apple Silicon via uv sync --extra mlx and EXPLORERBACKEND=mlx on port 8791, with a PyTorch backend (MPS/CUDA/CPU) and auto backend selection also available.
More from Research
- Ruliobiology: interactive slides grow creatures from tiny cellular automata rules — AjdDavison · 2026-10-27
- OpenAI internal model claims to refute Exponential Time Hypothesis; independent review organized — Singularitarian · 2026-10-27
- Music Lenses: ISMIR paper uses sparse autoencoders for steerable concept-level music retrieval — umpedronosapato · 2026-10-27
- Neuronpedia adds 7M+ neurons for Qwen3-32B, Gemma and Llama 3.3 70B to its interpretability encyclopedia — aryaman2020 · 2026-10-27
- Matthew Green hails stunning new ETH result pending Lean proof validity — matthew_d_green · 2026-10-27
- Heuristic for prediction accuracy: a bimodal probability histogram peaking at 0 and 1 — moultano · 2026-10-27