Lightlines
By orchestra-research

nnsight-remote-interpretability

orchestra-research887 installs

Probe large neural networks without local GPUs.

What it does

Guides experiments that inspect and alter neural-network activations using nnsight, with optional remote execution on large models.

When to use it

Use for interpretability experiments on models too large for local GPUs, or for accessing and intervening in PyTorch model internals.

How to use it

It changes how the agent implements neural-network interpretability experiments, using tracing contexts and explicitly saved activations.

What you provide

  • An existing project

Uses

Latest review

What you need · 3

The declared dependency is nnsight version 0.5.0 or newer.

The declared dependency is torch version 2.0.0 or newer.


Access · 2

This skill

NDIF

Execute

Model experiments on NDIF

Write

Sends experiment prompts to NDIF

nnsight

Execute

Installs nnsight