Open ngreenwald opened 4 years ago
I think this would be a nice feature to add to coincide with versioning the data. This would allow us to specify which version of the model was used to create what. A bonus would be to select different versions of the model, but at a minimum just showing a version number would suffice.
Also we could think about what a versioning system would look like beyond the sequential numbering we're currently using for model weights.
Unfortunately, the integer versioning is a requirement of TensorFlow Serving.
Currently, users have no way of knowing what version of a specific model they're using, or if it has been changed recently. It would be good to provide some basic metadata on what model is currently deployed to allow users to track what software produced their results