Hey Folks, I’m Dominik! I’m an active contributor to the Kubeflow project, for example, I was the Release Manager of the last 1.7 release. I was browsing a bit on the Flyte website and saw this interesting comparison between Flyte and Kubeflow. This is of course a comparison between Kubeflow Pipelines and Flyte, with pipelines being the component most similar to Flyte. I think the article is great and it definitely points out some of the shortcomings of pipelines. But as most of you know pipelines is just one component of Kubeflow, we have training-operator, katib, the notebook controller and all of the different Kubeflow UIs. I was wondering if the community has any interest in doing some kind of collaboration between Kubeflow and Flyte. We’ve been lately looking more into this as we believe that having an interoperable open source MLOps ecosystem benefits all of us. For the last release we did a really cool collaboration with BentoML to provide an additional serving option besides KServe. By browsing some GitHub issues I saw that there is at least some interest in integrating with Kubeflow’s training operator, so I thought I might just ask if there might be any interest in doing anything further with the whole Kubeflow project. Anyways just wanted to reach out in case anyone is interested and say that you are doing an awesome job with this project!

@Dominik Fleischmann firstly welcome and you are absolutely right the comparison is about Kfp and potentially as of a date. Now for training operator Flyte already works with training operator and infact many folks use it that way. Here are the docs https://docs.flyte.org/projects/cookbook/en/latest/integrations.html#native-backend-plugins If you notice they are called kubeflow pytorch/tensorflow/training operator as these were done earlier but since been migrated to use training operator If I may voice the community we will be more than ecstatic to collaborate and coordinate

Hey Ketan! That is awesome to hear. I don’t think I understand the last sentence

Also Flyte is from Atila focused on pipelines

. What exactly do you mean by that?

Typo fixed it

Perfect that is precisely why I was reaching out. While Kubeflow Pipelines is obviously a first class citizen I don’t think users should be limited to it if they want to leverage the other features of Kubeflow. And Flyte might benefit from being able to integrate with a more end 2 end platform.

Absolutely 🙌 ❤️

This is great thank you @David Espejo (he/him) !