Transformers.js Attempts to Improve Browser Cache Efficiency

Currently, when running AI models in the browser using Transformers.js, calling the same model from different domains (Origins) causes inefficiency where the model must be downloaded again every time, even though it has already been cached.

For example, when using the Whisper model, if the domain differs, about 177MB of data must be redundantly downloaded and stored.

This article shares the results of an experiment testing whether the Cross-Origin Storage API, proposed by Thomas Steiner of the Google Chrome team, can be used to share model resources across different domains, thereby improving cache efficiency and enhancing user experience.