You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This project simplifies MLOps in Kubernetes by providing a Terraform module which creates a Kubeflow instance.
23
+
This project simplifies MLOps in Kubernetes by providing Kubeflow in Helm and Terraform package formats. This allows scaling Kubeflow usage with the rest of your production systems.
24
24
25
25
Kubeflow provides a cloud-native AI platform which can be used to deploy applications in
26
26
scientific computing, traditional machine learning, and generative AI.
@@ -32,9 +32,15 @@ This module is primarily focussed on the Jupyter notebook environment initially
32
32
33
33
## Architecture
34
34
35
-
This module is built on top of the official [Kubeflow Manifests repo](https://github.com/kubeflow/manifests) which contains _Kustomizations_ for the various components of Kubeflow.
35
+
This system is built on top of the official [Kubeflow Manifests repo](https://github.com/kubeflow/manifests) which contains _Kustomizations_ for the various components of Kubeflow.
36
+
37
+
We provide a terraform and helm-based interface for managing Kubeflow via GitOps. Because Kubeflow is a collection of modular components, this project relies on ArgoCD for combining them.
36
38
37
-
We provide a terraform and helm-based interface for managing Kubeflow via GitOps
39
+
### Design Tenets
40
+
41
+
1. Integrate with production systems that already use Terraform and Helm
42
+
2. Embrace GitOps for Kubernetes resources on the popular ArgoCD project
43
+
3. Enable adoption of cloud-native/AI tools beyond the scope of Kubeflow (e.g. Ray, MLFlow)
@@ -131,48 +137,3 @@ Moving the deployment between different states of configuration can be challengi
131
137
due to the dependencies between components in the cluster.
132
138
133
139
If you have made a change to a dependency such as istio, or an auth component such as dex, it can be a good idea to re-create pods such that they re-initialise. This can be done by scaling to 0 then back up again, or simply deleting a pod managed by a deployment.
| <aname="input_bootstrap_set"></a> [bootstrap\_set](#input\_bootstrap\_set)| Value block with custom STRING values to be merged with the values yaml. | <pre>list(object({<br> name = string<br> value = string<br> }))</pre> |`null`| no |
168
-
| <aname="input_bootstrap_set_sensitive"></a> [bootstrap\_set\_sensitive](#input\_bootstrap\_set\_sensitive)| Value block with custom sensitive values to be merged with the values yaml that won't be exposed in the plan's diff. | <pre>list(object({<br> path = string<br> value = string<br> }))</pre> |`null`| no |
169
-
| <aname="input_bootstrap_values"></a> [bootstrap\_values](#input\_bootstrap\_values)| Extra values |`list(string)`|`[]`| no |
170
-
| <aname="input_enable_argocd"></a> [enable\_argocd](#input\_enable\_argocd)| n/a |`bool`|`true`| no |
171
-
172
-
## Outputs
173
-
174
-
No outputs.
175
-
176
-
## Not sure where to start?
177
-
178
-
Open an issue or message us directly in [Discord](https://discord.gg/QFjCpMjqRY).
0 commit comments