A 32,768-feature Top-K SAE trained on the layer 14 residual stream of
Qwen3 1.7B Base using 500M tokens from FineWeb-Edu. Includes checkpoint, 100M tokens of recorded activations, and feature interpretations.
Browse the features at
nanointerpret.pages.dev, or run the visualizer locally:
1git clone https://github.com/Belluxx/nanointerpret
2cd nanointerpret
3pip install -r requirements.txt
4hf download Belluxx/nanointerpret-qwen3 --local-dir artifacts/nanointerpret-qwen3
5python visualize.py --activations artifacts/nanointerpret-qwen3/activations
Training code and details:
Belluxx/nanointerpret.
