Hf Model Registry

The problem

Manually managing large machine learning artifacts via browser or git is slow and prone to corruption. Developers struggle with interrupted downloads, disk space management, and inconsistent metadata when publishing models.

What it does

• Authenticates and manages identity across Hugging Face organizations.
• Searches the Hub for specific model weights, datasets, and Spaces.
• Downloads large checkpoints with built-in resumability and revision pinning.
• Queries dataset schemas and previews data using SQL over Parquet files.
• Uploads trained artifacts with automated repository creation and tagging.

Frameworks & tools

Hugging Face CLI (hf), PyTorch, TensorFlow, Transformers, Datasets, and Parquet.

Why this beats prompting it yourself

This skill handles the nuances of the new 'hf' CLI syntax, replacing legacy commands. It ensures you use optimized transfer methods like resumable large-folder uploads rather than basic git clones which fail on large binary files.

Use cases

• Fetch specific model revisions for reproducible training pipelines.
• Preview dataset rows via SQL to verify data quality before downloading gigabytes.
• Automate the publication of fine-tuned LoRA adapters with proper README cards.
• Clean up local Hub caches to recover disk space on compute instances.

Known limitations

Requires the new Hugging Face CLI (hf) to be installed; not intended for in-training data streaming.