training-runner

This skill generates and executes robust pretraining pipelines for AI models by creating scripts that manage the entire lifecycle. It supports frameworks like NanoGPT, LitGPT, and Hugging Face Trainer while enabling background execution and checkpoint-safe recovery.

Yes, you can run Hugging Face Trainer in the background with live monitoring. The skill provides multiple execution patterns including background execution, delivering real-time insights into training progress and automatically handling interruptions or divergence.

Yes, training-runner supports multi-phase training strategies and checkpoint recovery. It enables checkpoint-safe recovery to automatically handle interruptions, ensuring your pretraining pipeline can resume reliably without losing progress.

The best way to track GPU-hours and cost estimation during model development is using this skill's integrated cost tracking feature. It automatically estimates and tracks the GPU-hours consumed throughout your pretraining pipeline execution.

You can execute a NanoGPT pretraining pipeline with checkpoint management by using this skill to generate the necessary scripts. It handles output management for training artifacts and ensures checkpoint-safe recovery for long-running processes.