Regarding replacing a Hub model

[sayakpaul]

I am writing to know what's the best way to replace a model on TensorFlow Hub? Today, I noticed the int8 EAST model is a bit buggy and I worked on a fix today. I wish to replace the existing one now. What should be the course of action here?

[arnoegw]

To fix tfhub.dev/sayakpaul/lite-model/east-text-detector/int8/1, you can publish a new model version tfhub.dev/sayakpaul/lite-model/east-text-detector/int8/2 (or whichever higher integer makes sense). Both versions will coexist (each with its own documentation), but users who browse to model version /1 will see a notice pointing out the newer version.

For an existing example, see https://tfhub.dev/google/lite-model/aiy/vision/classifier/plants_V1/1 and its updates to /2 and /3. In that example, the authors chose to put a Changelog into the documentation that grows with each version and explains what improvement each version has brought.

In documentation, you can use an unversioned link https://tfhub.dev/sayakpaul/lite-model/east-text-detector/int8 to send web browsers to the most recent model version. However, model handles uses in code must be versioned, to avoid unexpected changes of behavior and race conditions in distributed training.

[sayakpaul]

Thanks @arnoegw. Will proceed accordingly.