training-runner

What problem does it solve?

This Skill streamlines the complex process of setting up, executing, and monitoring large-scale model pretraining, ensuring robustness and efficient resource utilization.

Core Features & Use Cases

• End-to-End Training Execution: Manages the entire lifecycle of model pretraining, from script generation to checkpoint management.
• Framework Agnostic: Supports popular training frameworks like NanoGPT, LitGPT, and Hugging Face Trainer.
• Reliable Execution: Offers multiple execution patterns (SSH, Kaggle, background) with built-in support for long-running processes and recovery.
• Live Monitoring & Recovery: Provides real-time insights into training progress and automatically handles interruptions and divergence.
• Cost Tracking: Integrates cost estimation and tracking for GPU-hours consumed.

Quick Start

Use the training-runner skill to generate and execute a NanoGPT pretraining pipeline with background execution and live monitoring.