Hugging Face Tests New Cross-Origin Storage API in Transformers.js
Hugging Face is experimenting with a proposed web API called Cross-Origin Storage, allowing web-based AI models running in Transformers.js to securely share data like model weights or inference cache across different websites. This could reduce redundant downloads and make AI tools faster and more seamless across multiple platforms.
Hugging Face released an experimental version of Transformers.js with support for the proposed Cross-Origin Storage API. This new feature allows web applications to securely share data across different websites, even when they come from different domains. In plain English, it means your browser can safely store and access large AI model files or cached computations from multiple AI tools without re-downloading them each time.
This matters because it could make using AI tools across different websites much more efficient. Instead of each website needing to download the same multi-gigabyte AI model independently, they could share a single cached copy. Imagine being able to use a writing assistant on one site and then seamlessly access its language model data on another without waiting for another download. It simplifies the process and makes AI tools more integrated into your daily browsing experience, all while maintaining security boundaries between sites.
Hugging Face has published a technical blog post detailing their experiments with the API, including code examples and performance benchmarks. If you're curious, you can try this out today by visiting the Hugging Face blog and checking out their experimental Transformers.js implementation. Look for the section on Cross-Origin Storage and follow the instructions to test it out in a safe environment.