Hi everyone, Does anyone have experience in making Flyte working with GPU workers that are on a cloud provider different from GCP, AWS? Flyte seems to be great workflow orchestration tool for K8S ML. Recently, my company has more needs for GPUs and the GCP availability has not been good (we can't get GPUs). So we've looked into other platforms that provide much better GCP availability (e.g., modal.com, https://nebius.com/). I'm wondering if Flyte can run workloads in these platforms.

As long as it sits on a Kubernetes deployment it shouldn’t matter what the underlying cloud provider is, right?

In theory that's true. But these neo clouds seem to not be directly compatible with k8s

This seems to suggest they do provide a managed Kubernetes setup, akin to e.g. EKS: https://nebius.com/services/managed-kubernetes (disclaimer: I have no first-hand experience with it)

Union has a partnership with Nebius. https://www.union.ai/blog-post/union-ai-on-nebius-orchestrating-the-future-of-ai-workloads-in-the-cloud Since flyte is the common backend, i presume this is possible.

@little-library-90787 Union can schedule across nebius and GCP we already do that for some customers