karpathy

Iterate training script changes to minimize a scalar metric.

150|19|Updated Jun 15, 2026

Search Tags:#machine learning #AI optimization #code modification #model tuning #autoresearch

Ready to run

npx skills add https://github.com/gaasher/Agent-Loop-Skills --skill karpathy-gaasher

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: karpathy
Source: https://github.com/gaasher/Agent-Loop-Skills/tree/main/loops/karpathy
Command: npx skills add https://github.com/gaasher/Agent-Loop-Skills --skill karpathy-gaasher

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve?

Karpathy Autoresearch addresses the need for a hands-off AI-driven loop that iterates through changes to a training script to optimize a single scalar metric, such as validation accuracy.

Core Features & Use Cases

Autonomous Research: Automates the process of hacking training code, running it, and deciding which changes to keep based on metric improvements.
Iterative Optimization: Continuously improves upon the training script to achieve the lowest scalar metric.
Use Case: For users who want to optimize a model without manual intervention, such as in machine learning research or model refinement.

Quick Start

Set up the environment with loop.run.yaml and initiate the experiment by running the uv run train.py command.

Frequently Asked Questions about karpathy

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

FAQPage Schema

How do I automate machine learning model tuning without manual code changes?▼

Automated model tuning is achieved through an autonomous AI research loop that iteratively hacks training scripts, runs them, and keeps changes that minimize a scalar metric like validation accuracy.

What is autonomous AI research for iterative code optimization?▼

Autonomous AI research is a hands-off loop that continuously modifies and runs training code to optimize a single scalar metric, automatically deciding which changes to retain based on measurable improvements.

Can I use this autoresearch loop for my own training environment?▼

Yes, this autoresearch loop requires a user-provided training environment and Python 3.9 or higher, allowing you to integrate the automated code modification process into your existing machine learning setup.

How do I set up an autonomous AI optimization loop for machine learning?▼

To set up the autonomous optimization loop, configure your environment using the loop.run.yaml file and initiate the experiment by executing the uv run train.py command in your terminal.

What are the limitations of using an autonomous AI loop for code modification?▼

The autonomous code modification loop is limited to optimizing a single scalar metric within a fixed budget, requiring a user-provided training environment and Python 3.9+ to execute the iterative process.

Why does automated model tuning focus on minimizing a single scalar metric?▼

Automated model tuning focuses on minimizing a single scalar metric because the autonomous loop uses that specific value to decide whether to keep or discard iterative training code changes within a fixed budget.