Separate charts [BREAKING CHANGE TO HELM CHARTS] (#220)

Split the model catalog in `charts/kubeai/` into its own chart
(`charts/models/`). Move CRDs to `templates/` to allow users to `helm
upgrade` CRDs.

Fixes #213

---------

Co-authored-by: Sam Stoelinga <[email protected]>

.gitignore

@@ -27,7 +27,10 @@ go.work
 *~
 
 # Files that might be committed from running guides
+/kubeai.yaml
+/kubeai-models.yaml
 /helm-values.yaml
+/model-helm-values.yaml
 Chart.lock
 
 # Ignore python virtual env

Makefile

@@ -45,8 +45,14 @@ help: ## Display this help.
 ##@ Development
 
 .PHONY: manifests
-manifests: controller-gen ## Generate CustomResourceDefinition objects.
-	$(CONTROLLER_GEN) crd webhook paths="./..." output:crd:artifacts:config=charts/kubeai/charts/crds/crds
+manifests: controller-gen yq
+	# Generate CustomResourceDefinition objects.
+	$(CONTROLLER_GEN) crd webhook paths="./..." output:crd:artifacts:config=charts/kubeai/templates/crds/
+
+	# Generate model manifests.
+	rm -f ./manifests/models/*
+	helm template ./charts/models --set all.enabled=true | \
+	$(YQ) -s '"./manifests/models/" + .metadata.name + ".yaml"' --no-doc
 
 .PHONY: generate
 generate: controller-gen ## Generate code containing DeepCopy, DeepCopyInto, and DeepCopyObject method implementations.
@@ -135,12 +141,14 @@ CONTROLLER_GEN ?= $(LOCALBIN)/controller-gen
 ENVTEST ?= $(LOCALBIN)/setup-envtest
 GOLANGCI_LINT = $(LOCALBIN)/golangci-lint
 CRD_REF_DOCS ?= $(LOCALBIN)/crd-ref-docs
+YQ ?= $(LOCALBIN)/yq
 
 ## Tool Versions
 KUSTOMIZE_VERSION ?= v5.4.2
 CONTROLLER_TOOLS_VERSION ?= v0.15.0
 ENVTEST_VERSION ?= release-0.18
 GOLANGCI_LINT_VERSION ?= v1.59.1
+YQ_VERSION ?= v4.44.3
 
 .PHONY: kustomize
 kustomize: $(KUSTOMIZE) ## Download kustomize locally if necessary.
@@ -162,6 +170,11 @@ golangci-lint: $(GOLANGCI_LINT) ## Download golangci-lint locally if necessary.
 $(GOLANGCI_LINT): $(LOCALBIN)
 	$(call go-install-tool,$(GOLANGCI_LINT),github.com/golangci/golangci-lint/cmd/golangci-lint,$(GOLANGCI_LINT_VERSION))
 
+.PHONY: yq
+yq: $(YQ) ## Download yq locally if necessary.
+$(YQ): $(LOCALBIN)
+	$(call go-install-tool,$(YQ),github.com/mikefarah/yq/v4,$(YQ_VERSION))
+
 .PHONY: docs
 docs: generate-kubernetes-api-reference
 	cd ./docs && \

charts/kubeai/Chart.yaml

@@ -21,18 +21,6 @@ dependencies:
   - name: openwebui
     condition: openwebui.enabled
     version: 0.1.1
-  # Custom Resource Definitions are cluster scoped. Sometimes cluster admins need to install
-  # these resources as a separate step.
-  # https://kubernetes.io/docs/concepts/extend-kubernetes/api-extension/custom-resources/
-  - name: crds
-    condition: crds.enabled
-    version: 0.1.1
-  # Models are the machine learning models that kubeai will serve.
-  # These are instances of the Model custom resource.
-  - name: models
-    condition: models.enabled
-    version: 0.1.1
-
 
 keywords: ["LLM", "AI"]
 
@@ -41,7 +29,7 @@ home: https://www.substratus.ai
 
 maintainers:
 - name: nstogner
-  email: nicholas.stogner@gmail.com
+  url: https://www.linkedin.com/in/nstogner/
 - name: samos123
   email: [email protected]
   url: https://www.linkedin.com/in/samstoelinga/

charts/kubeai/charts/models/Chart.yaml → charts/models/Chart.yaml

@@ -22,3 +22,10 @@ version: 0.1.1
 # follow Semantic Versioning. They should reflect the version the application is using.
 # It is recommended to use it with quotes.
 appVersion: "1.16.0"
+
+maintainers:
+- name: nstogner
+  url: https://www.linkedin.com/in/nstogner/
+- name: samos123
+  email: [email protected]
+  url: https://www.linkedin.com/in/samstoelinga/

charts/kubeai/charts/models/templates/models.yaml → charts/models/templates/models.yaml

@@ -1,5 +1,5 @@
 {{- range $name, $model := .Values.catalog}}
-{{- if $model.enabled}}
+{{- if or $model.enabled $.Values.all.enabled }}
 ---
 apiVersion: kubeai.org/v1
 kind: Model
@@ -14,11 +14,11 @@ spec:
   {{- end }}
   {{- with $model.args }}
   args:
-  {{ . | toYaml | nindent 4 }}
+  {{- toYaml . | nindent 4 }}
   {{- end }}
   {{- with $model.env }}
   env:
-  {{ . | toYaml | nindent 4 }}
+  {{- toYaml . | nindent 4 }}
   {{- end }}
   {{- with $model.minReplicas }}
   minReplicas: {{ . }}

charts/kubeai/charts/models/values.yaml → charts/models/values.yaml

@@ -1,3 +1,7 @@
+all:
+  # Enable all models instead of enabling them one-by-one via .catalog.<name>.enabled
+  enabled: false
+
 catalog:
   # Mistral #
   e5-mistral-7b-instruct-cpu:
@@ -79,4 +83,3 @@ catalog:
     url: "hf://Systran/faster-whisper-medium.en"
     engine: FasterWhisper
     resourceProfile: cpu:1
-

docs/README.md

@@ -58,21 +58,25 @@ helm repo update
 Install KubeAI and wait for all components to be ready (may take a minute).
 
 ```bash
-cat <<EOF > helm-values.yaml
-models:
-  catalog:
-    gemma2-2b-cpu:
-      enabled: true
-      minReplicas: 1
-    qwen2-500m-cpu:
-      enabled: true
-    nomic-embed-text-cpu:
-      enabled: true
+helm install kubeai kubeai/kubeai --wait --timeout 10m
+```
+
+Install some predefined models.
+
+```bash
+cat <<EOF > kubeai-models.yaml
+catalog:
+  gemma2-2b-cpu:
+    enabled: true
+    minReplicas: 1
+  qwen2-500m-cpu:
+    enabled: true
+  nomic-embed-text-cpu:
+    enabled: true
 EOF
 
-helm upgrade --install kubeai kubeai/kubeai \
-    -f ./helm-values.yaml \
-    --wait --timeout 10m
+helm install kubeai-models kubeai/models \
+    -f ./kubeai-models.yaml
 ```
 
 Before progressing to the next steps, start a watch on Pods in a standalone terminal to see how KubeAI deploys models. 

docs/how-to/configure-autoscaling.md

@@ -4,7 +4,7 @@ This guide with cover how to configure KubeAI [autoscaling](../concepts/autoscal
 
 ## System Settings
 
-KubeAI administrators can define system-wide autoscaling settings by setting the following helm values:
+KubeAI administrators can define system-wide autoscaling settings by setting the following Helm values (for the `kubeai/kubeai` chart):
 
 Example:
 
@@ -24,21 +24,20 @@ The following settings can be configured on a model-by-model basis.
 
 ### Model settings: helm
 
-If you are managing models via Helm, you can use:
+If you are managing models via the `kubeai/models` Helm chart, you can use:
 
 ```yaml
 # helm-values.yaml
-models:
-  catalog:
-    model-a:
-      # ...
-      minReplicas: 1
-      maxReplicas: 9
-      targetRequests: 250
-      scaleDownDelaySeconds: 45
-    model-b:
-      # ...
-      disableAutoscaling: true
+catalog:
+  model-a:
+    # ...
+    minReplicas: 1
+    maxReplicas: 9
+    targetRequests: 250
+    scaleDownDelaySeconds: 45
+  model-b:
+    # ...
+    disableAutoscaling: true
 # ...
 ```
 

docs/how-to/configure-speech-to-text.md

@@ -6,14 +6,14 @@ KubeAI provides a Speech to Text endpoint that can be used to transcribe audio f
 You can create new models by creating a Model CRD object or by enabling a model from the model catalog.
 
 ### Enable from model catalog
-KubeAI provides predefined models in the model catalog. To enable the Speech to Text model, you can set the `enabled` flag to `true` in the `helm-values.yaml` file.
+KubeAI provides predefined models in the `kubeai/models` Helm chart. To enable the Speech to Text model, you can set the `enabled` flag to `true` in your values file.
 
 ```yaml
-models:
-  catalog:
-    faster-whisper-medium-en-cpu:
-      enabled: true
-      minReplicas: 1
+# models-helm-values.yaml
+catalog:
+  faster-whisper-medium-en-cpu:
+    enabled: true
+    minReplicas: 1
 ```
 
 ### Enable by creating Model CRD

docs/how-to/install-models.md

@@ -4,43 +4,42 @@ This guide provides instructions on how to configure KubeAI models.
 
 ## Installing models with helm
 
+KubeAI provides a [chart](https://github.com/substratusai/kubeai/blob/main/charts/models) that contains preconfigured models.
+
 ### Preconfigured models with helm
 
-When you are defining KubeAI Helm values, you can install a preconfigured Model by setting `enabled: true`. You can view a list of all preconfigured models [here](https://github.com/substratusai/kubeai/blob/main/charts/kubeai/charts/models/values.yaml). NOTE: When you are installing the KubeAI chart, the catalog is accessed under `.models.catalog.<model-name>`:
+When you are defining Helm values for the `kubeai/models` chart you can install a preconfigured Model by setting `enabled: true`. You can view a list of all preconfigured models in the chart's [default values file](https://github.com/substratusai/kubeai/blob/main/charts/models/values.yaml). 
 
 ```yaml
 # helm-values.yaml
-models:
-  catalog:
-    llama-3.1-8b-instruct-fp8-l4:
-      enabled: true
+catalog:
+  llama-3.1-8b-instruct-fp8-l4:
+    enabled: true
 ```
 
 You can optionally override preconfigured settings, for example, `resourceProfile`:
 
 ```yaml
 # helm-values.yaml
-models:
-  catalog:
-    llama-3.1-8b-instruct-fp8-l4:
-      enabled: true
-      resourceProfile: nvidia-gpu-l4:2 # Require "2 NVIDIA L4 GPUs"
+catalog:
+  llama-3.1-8b-instruct-fp8-l4:
+    enabled: true
+    resourceProfile: nvidia-gpu-l4:2 # Require "2 NVIDIA L4 GPUs"
 ```
 
 ### Custom models with helm
 
-If you prefer to add a custom model via the same Helm chart you use for installed KubeAI, you can add your custom model entry into the `.models.catalog` array of your existing Helm values file:
+If you prefer to add a custom model via the same Helm chart you use for installed KubeAI, you can add your custom model entry into the `.catalog` array of your existing values file for the `kubeai/models` Helm chart:
 
 ```yaml
 # helm-values.yaml
-models:
-  catalog:
-    my-custom-model-name:
-      enabled: true
-      features: ["TextEmbedding"]
-      owner: me
-      url: "hf://me/my-custom-model"
-      resourceProfile: CPU:1
+catalog:
+  my-custom-model-name:
+    enabled: true
+    features: ["TextEmbedding"]
+    owner: me
+    url: "hf://me/my-custom-model"
+    resourceProfile: CPU:1
 ```
 
 ## Installing models with kubectl