transformer-lens-interpretability
Provides guidance for mechanistic interpretability research using TransformerLens to inspect and manipulate transformer internals via HookPoints and activation caching. Use when reverse-engineering model algorithms, studying attention patterns, or performing activation patching experiments.
- Score
- Unranked
- Evidence
- No signals yet
- Last commit
- Not known
- Listed
Install
armory install transformer-lens-interpretability --cli claudewrites the skill into.claude/skills/transformer-lens-interpretability/SKILL.mdListed as compatible
# fetches the source and writes it to:
.claude/skills/transformer-lens-interpretability/SKILL.mdNeeds the armory CLI · not on npm yet, build it from cli/ in the repository
What it is
Provides guidance for mechanistic interpretability research using TransformerLens to inspect and manipulate transformer internals via HookPoints and activation caching. Use when reverse-engineering model algorithms, studying attention patterns, or performing activation patching experiments.
When to use it
Provides guidance for mechanistic interpretability research using TransformerLens to inspect and manipulate transformer internals via HookPoints and activation caching. Use when reverse-engineering model algorithms, studying attention patterns, or performing activation patching experiments.
How to install / invoke
# Copy the skill into your .claude/skills/ directory
curl -sL https://raw.githubusercontent.com/davila7/claude-code-templates/main/cli-tool/components/skills/ai-research/mechanistic-interpretability-transformer-lens/SKILL.md -o .claude/skills/mechanistic-interpretability-transformer-lens/SKILL.md
Notes
Extracted from davila7/claude-code-templates, mechanistic-interpretability-transformer-lens category.