Karthik Ramasamy

Distirbuted trianing TF (1.x and 2.0) > tfjob deployer is working fine (have an integration test). Didn't test it using tf 2.0 yet. Planning to add examples by end of...

We have no cycles to do this in 0.6 so removed it from 0.6 board.

>Would it be better to output it into a temporary directory? Yes that would be better. Related issue: #75

>create a fake virtualenv, run pip install within new virtualenv, and package that up. This might be running under OS/env that might work with the base container right? >ignore requirements.txt...

no updates. I should be able to look into it next week.

you have to install the requirements in https://github.com/kubeflow/fairing/blob/master/examples/prediction/requirements.txt . We didn't create a separate image for this example. We will add "!pip install -r requirements.txt" as the first cell in...

Right now users can do arbitrary python predict function which supports TF models but it is not an efficient method. Ref: https://github.com/kubeflow/fairing/tree/master/examples/prediction This issue is specially for using TF serving...

This will be fixed once KFServing support is fully done. Related: #299

CC @martin-gorner

CC @richardsliu @abhi-g