generate

substratusai · Dec 5, 2024 · 8c177b5 · 8c177b5
1 parent 3f4fcf8
commit 8c177b5
Show file tree

Hide file tree

Showing 2 changed files with 10 additions and 4 deletions.
diff --git a/charts/kubeai/templates/crds/kubeai.org_models.yaml b/charts/kubeai/templates/crds/kubeai.org_models.yaml
@@ -165,6 +165,8 @@ spec:
 
 
                   "hf://<repo>/<model>"
+                  "pvc://<pvcName>"
+                  "pvc://<pvcName>/<pvcSubpath>"
                   "gs://<bucket>/<path>" (only with cacheProfile)
                   "oss://<bucket>/<path>" (only with cacheProfile)
                   "s3://<bucket>/<path>" (only with cacheProfile)
@@ -191,15 +193,19 @@ spec:
             type: object
             x-kubernetes-validations:
             - message: cacheProfile is only supported with urls of format "hf://...",
-                "s3://...", "gs://...", or "oss://..." at the moment.
-              rule: '!has(self.cacheProfile) || self.url.startsWith("hf://") || self.url.startsWith("s3://")
-                || self.url.startsWith("gs://") || self.url.startsWith("oss://")'
+                "pvc://...", "s3://...", "gs://...", or "oss://..." at the moment.
+              rule: '!has(self.cacheProfile) || self.url.startsWith("hf://") || self.url.startsWith("pvc://")
+                || self.url.startsWith("s3://") || self.url.startsWith("gs://") ||
+                self.url.startsWith("oss://")'
             - message: urls of format "s3://..." only supported when using a cacheProfile
               rule: '!self.url.startsWith("s3://") || has(self.cacheProfile)'
             - message: urls of format "gs://..." only supported when using a cacheProfile
               rule: '!self.url.startsWith("gs://") || has(self.cacheProfile)'
             - message: urls of format "oss://..." only supported when using a cacheProfile
               rule: '!self.url.startsWith("oss://") || has(self.cacheProfile)'
+            - message: urls of format "pvc://..." do not support cacheProfile. Remove
+                cacheProfile or use a different url.
+              rule: self.url.startsWith("pvc://") || has(self.cacheProfile)
             - message: minReplicas should be less than or equal to maxReplicas.
               rule: '!has(self.maxReplicas) || self.minReplicas <= self.maxReplicas'
             - message: adapters only supported with VLLM engine.

diff --git a/docs/reference/kubernetes-api.md b/docs/reference/kubernetes-api.md
@@ -76,7 +76,7 @@ _Appears in:_
 
 | Field | Description | Default | Validation |
 | --- | --- | --- | --- |
-| `url` _string_ | URL of the model to be served.<br />Currently the following formats are supported:<br /><br />For VLLM, FasterWhisper, Infinity engines:<br /><br />"hf://<repo>/<model>"<br />"gs://<bucket>/<path>" (only with cacheProfile)<br />"oss://<bucket>/<path>" (only with cacheProfile)<br />"s3://<bucket>/<path>" (only with cacheProfile)<br /><br />For OLlama engine:<br /><br />"ollama://<model>" |  | Required: \{\} <br /> |
+| `url` _string_ | URL of the model to be served.<br />Currently the following formats are supported:<br /><br />For VLLM, FasterWhisper, Infinity engines:<br /><br />"hf://<repo>/<model>"<br />"pvc://<pvcName>"<br />"pvc://<pvcName>/<pvcSubpath>"<br />"gs://<bucket>/<path>" (only with cacheProfile)<br />"oss://<bucket>/<path>" (only with cacheProfile)<br />"s3://<bucket>/<path>" (only with cacheProfile)<br /><br />For OLlama engine:<br /><br />"ollama://<model>" |  | Required: \{\} <br /> |
 | `adapters` _[Adapter](#adapter) array_ |  |  |  |
 | `features` _[ModelFeature](#modelfeature) array_ | Features that the model supports.<br />Dictates the APIs that are available for the model. |  | Enum: [TextGeneration TextEmbedding SpeechToText] <br /> |
 | `engine` _string_ | Engine to be used for the server process. |  | Enum: [OLlama VLLM FasterWhisper Infinity] <br />Required: \{\} <br /> |