Hi 👋 Ive been using flyte to launch TFJobs on k8s using the

flytekit-kf-tensorflow

plugin. I noticed the

TFJob

task config doesn't contain an element for

evaluators

which is part of the

TFJob

spec. Does anyone have a work around for this? The use case is to have a data service running and use the evlauators section in the TF_CONFIG to set the data service worker information

Hi @Alykhan Tejani great to meet you. Evaluators was probably added later in TFJob and not supported by the plugin today. You can contribute it if you are open to it https://github.com/flyteorg/flyteplugins/blob/552f145c4632944a23e7e2d9fa2b2f065c4b14b9/go/tasks/plugins/k8s/kfoperators/tensorflow/tensorflow.go#L66 - else file an issue. @Fabio Grätz do you use it. Or @Byron Hsu

Thanks

Hey, no, torch only for us, sorry 😕

We use tf but don’t use evaluators. It can be easily added to the plugins. Or @L godlike @Yi Chiu are you interested to help?

I can help after I finish the current feature. Currently, I am working on the sync plugin interface and the ChatGPT plugin. Please give me a chance, thanks a lot!

great

@Alykhan Tejani Hello, I just finished the issue. Can you test it in your environment for me? If possible, can you also provide the screenshot of the evaluator? Here is the issue link and the code you should modify. Issue link: https://github.com/flyteorg/flyte/issues/4167 flyteidl and flyteplugins: https://github.com/flyteorg/flyte/pull/4168 flytekit: https://github.com/flyteorg/flytekit/pull/1870 Thanks a lot!