Ability to run on private custom models?

google-research / task_adaptation

Apache License 2.0

311 stars 35 forks source link

Ability to run on private custom models? #3

Closed dkossnick-figma closed 5 years ago

dkossnick-figma commented 5 years ago

I understand this repo requires the input model to be hosted on tensorflow hub. It seems like all tensorflow hub models must also be public. Is there any way to run this on private hosted or local models? When working on new model architectures, it would be great to be able to use this as an evaluation tool. Thank you!

akolesnikoff commented 5 years ago

Not necessarily: the input model can be a local model that is saved as hub module. It is generally easy to export any trained tensorflow model as hub model, see https://www.tensorflow.org/hub/tf1_hub_module#creating_a_new_module as a starting point.

Feel free to reopen this issue if you have further questions.