Hugging Face Releases Swift Client

Hugging Face announced a new Swift package called swift-huggingface. This package provides a complete client for the Hugging Face Hub, and will eventually replace the existing HubApi implementation in swift-transformers.

It focuses on solving the slow and unstable downloads, cache inconsistencies with the Python ecosystem, and complex authentication issues that have been experienced in the existing Swift environment.

Key features:

• Complete Hub API coverage: Supports all major Hub features including models, datasets, spaces, collections, and discussions.
• Robust file operations: Provides download progress tracking and a resume feature for restarting interrupted downloads.
• Python-compatible cache: Uses the same cache structure as the Python transformers library, allowing models already downloaded in Python to be reused in Swift apps.
• Flexible authentication and OAuth support: Through the TokenProvider pattern, environment variables, Keychain, and OAuth 2.0 can be explicitly managed, making it easy to build user-based apps.
• Xet storage support planned: Will dramatically increase download speeds through chunk-based deduplication in the future.