argoflow VS manifests

I have a couple of VM's running a Kubernetes cluster, I've been trying for a while to deploy Kubeflow on ArgoCD using 'ArgoFlow' (https://github.com/argoflow/argoflow), I can get ArgoFlow running on a Load Balancer and access the UI. I had a lot of trouble getting their script to work so I manually updated the .yaml files in my cloned Git repo to point to itself, when I deploy Kubeflow from their manifests, most things seem to fail with the error: "one or more synchronization tasks are not valid". Is there a different/easier way of deploying Kubeflow on my local cluster using ArgoCD? Been attempting this for months now.

Thank you for the reply, I did not know about Argo Workflows. What I've been trying to do is follow the https://github.com/argoflow/argoflow instructions to get it up and running.

Wouldn’t it make more sense to provision resources inside Kubernetes using specific tooling like ArgoCD or Flux? https://github.com/argoflow/argoflow

I have a baremetal k8s with Anthos and successfully installed ASM (Cloud Run for Anthos) on it. However, there is some conflict when trying to install Kubeflow following this repo (https://github.com/kubeflow/manifests).

That said I personally use Kubeflow hosted on a local baremetal kubernetes cluster (8 nodes, 4 gpus), but a lot of it is a bit of a bear to get installed correctly in a multi-machine environment (specifically this issue is still open and exposing the built-in dashboards outside of the cluster is a problem). Also because it's a Google product it's very clearly intended to run in the cloud with self-hosting being very much an afterthought

kind create cluster and then use the single installation command from https://github.com/kubeflow/manifests

I have deployed kubelfow v1.6 using using ArgoCD and it’s fairly simple. Every component of kubeflow has kustomze file ready kubeflow kustomize link. You just need to make argocd app for every component and then apply in that order.

For context, I run a local baremetal k8s cluster distributed over a number of machines and I've tried both [wandb](https://wandb.ai/site) and [kubeflow](https://www.kubeflow.org/), finding them to be a serious headache to manage in a local deployment. Almost none of the self-hosted builds they provide work out of the box, there's a frequent issues that have caused me significant amounts of data loss, and there are several known issues on both projects that have been open for years that make for my particular use-case difficult (e.g. [access outside of the cluster requiring a bunch of yak shaving](https://github.com/kubeflow/manifests/issues/974)).