[Feature request] ML Kit Entity Extraction dynamic model download not usable in mainland China – request for offline/bundled or self-hosted model support

Hello ML Kit team,

When using Entity Extraction with the dynamic download mode, the model files are fetched from Google servers at runtime.
However, due to network restrictions in mainland China, these model downloads often fail or timeout, making the feature unusable in production apps for users in this region.

Problem:

Dynamic model download cannot reliably complete inside mainland China because Google services are blocked or unstable.

Without a successful download, the Entity Extraction API cannot run at all.

Request

To make ML Kit more accessible for developers targeting China, we kindly request:

Bundled model support – allow developers to include the model in the app package (like an on-device static model) so no runtime download is required.

Custom hosting option – allow developers to download the model file in advance and host it on their own server/CDN, then load it into ML Kit at runtime.

Either of these options would let developers comply with local network conditions while still using ML Kit’s Entity Extraction.

Thank you for considering this improvement.