Teach LLMs to use NNSight with Skills
This repository also configures its own agents. See what skills tells them →
Plugin Claude Code
Bundles 19 skills · 2,479 tokens together
Neural network interpretability with nnsight 0.8 - tracing, intervening, patching, steering, debugging, and remote execution on NDIF.
Plugin Claude Code
Bundles 4 skills · 768 tokens together
Self-hosting NDIF, the server behind nnsight's remote=True - standing one up with docker run, compose, or the ndif CLI; deploying and sizing models; troubleshooting; and changing the server itself.
Plugin Claude Code
Bundles 3 skills · 462 tokens together
Architecture-agnostic interpretability with nnterp on nnsight 0.8 - one module vocabulary and standard values (layeroutput, attentionoutput, attentionprobabilities, ...) across 98 transformer families, the vision side of vision-language wrappers (model.vision, the tower blocks, the projector, imagetokenmask, image.