skills.rest
cesaregarza
cesaregarzaCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

mechinterp-runner

Execute mechanistic interpretability experiments from JSON specs and emit structured results.

1|Updated Jul 9, 2024
Search Tags:#CLI#reproducibility#experiments#interpretability#mechinterp#JSON-specs#SplatNLP
One-click install
npx skills add https://github.com/cesaregarza/SplatNLP --skill mechinterp-runner
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: mechinterp-runner
Source: https://github.com/cesaregarza/SplatNLP/tree/main/.claude/skills/mechinterp-runner
Command: npx skills add https://github.com/cesaregarza/SplatNLP --skill mechinterp-runner

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill orchestrates mechanistic interpretability experiments by loading JSON specs, routing tasks to the appropriate experiment runners, and producing structured results for reproducibility and analysis.

Core Features & Use Cases

  • Loads and validates experiment specs from JSON files and CLI prompts.
  • Routes to the correct runner type (family sweeps, itemsets, interactions, minimal cores, validation) and executes analyses with constraint enforcement.
  • Emits structured JSON results with diagnostics for downstream analysis and reporting.

Quick Start

Use the mechinterp-runner to execute a simple experiment by supplying a JSON spec via the CLI, for example: poetry run python -m splatnlp.mechinterp.cli.runner_cli family-sweep --feature-id 6235 --family quick_respawn --model ultra

Frequently Asked Questions about mechinterp-runner

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run mechanistic interpretability experiments reproducibly from a centralized spec?▼

Mechanistic interpretability experiments can be run reproducibly by loading JSON specs and routing them to appropriate experiment runners. This orchestrator validates the spec, enforces constraints, and emits structured JSON results with diagnostics.

What types of mechanistic interpretability analyses can I execute using JSON specs?▼

JSON specs support family sweeps, itemsets, interactions, minimal cores, and validation analyses. The orchestrator routes each spec to the correct runner type and executes the analysis via defined pipelines.

How do I execute a family sweep experiment via the CLI?▼

Execute a family sweep by supplying a JSON spec via the CLI, for example: poetry run python -m splatnlp.mechinterp.cli.runner_cli family-sweep --feature-id 6235 --family quick_respawn --model ultra.

Do I need any external dependencies to orchestrate mechinterp experiments from JSON specs?▼

No external dependencies are required to orchestrate mechinterp experiments from JSON specs. The runner validates and routes tasks internally, requiring only a valid JSON spec supplied via the CLI to execute.

What output format does the mechinterp experiment runner produce for downstream analysis?▼

The runner produces structured JSON results with diagnostics for downstream analysis and reporting. These outputs capture the experiment execution details and validation results in a centralized format.

Can I use this orchestrator for validation analyses on mechanistic interpretability models?▼

Yes, validation analyses are fully supported. The orchestrator routes JSON specs to the validation runner type, enforces constraints during execution, and emits structured results to validate mechanistic interpretability models.

Related Skills

Explore More Data & Analytics Skills→
goodfire-aigoodfire-ai
108

plan-experiment

Create structured experiment plans with analysis DAGs and pre-flight gates.

Official
Advanced
aliyunaliyun
213

alibabacloud-governance-evaluation-report

Query Alibaba Cloud Governance Center maturity data and output JSON risk reports.

Official
Advanced
yashpalsince2004yashpalsince2004

coreml

Integrate Core ML models into iOS apps for on-device inference with Swift.

Community
Advanced
jordanhubbardjordanhubbard
9

obliteratus

Remove refusal behaviors from open-weight LLMs using mechanistic interpretability techniques.

Community
Advanced