The Agent Skills Directory

[SAFE]: The skill serves as a collection of machine learning training templates. All code provided (PyTorch Lightning modules, Hugging Face Trainer configurations, and custom training loops) is standard and aligns with established best practices in the field.
[SAFE]: Dependencies identified (such as torch, pytorch_lightning, transformers, and wandb) are well-known, reputable libraries in the data science and machine learning ecosystem.
[SAFE]: File system operations are limited to saving and loading model checkpoints and training logs in local subdirectories (checkpoints/, outputs/).
[SAFE]: Network activity is restricted to wandb (Weights & Biases), a well-known service for machine learning experiment tracking and metrics visualization.

training-patterns