Skip to content

Commit

Permalink
Add OVMS OOTB with GPU support (#1262)
Browse files Browse the repository at this point in the history
* Add openvino with GPU support

* Update display name of gpu

* Use one image for both OOTB
  • Loading branch information
lucferbux authored May 19, 2023
1 parent 40cd8a4 commit 5d92707
Show file tree
Hide file tree
Showing 3 changed files with 66 additions and 4 deletions.
9 changes: 5 additions & 4 deletions manifests/modelserving/kustomization.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -4,8 +4,9 @@ commonLabels:
app: odh-dashboard
app.kubernetes.io/part-of: odh-dashboard
resources:
- servingruntimes-template.yaml
- ovms-ootb.yaml
- ovms-gpu-ootb.yaml
images:
- name: ovms-1
newName: quay.io/opendatahub/openvino_model_server
digest: sha256:20dbfbaf53d1afbd47c612d953984238cb0e207972ed544a5ea662c2404f276d
- name: ovms-1

This comment has been minimized.

Copy link
@shalberd

shalberd Jun 3, 2023

Contributor

@andrewballantyne @lucferbux interesting comment on golang-yaml and indentation of lists with two spaces before the hypens compared to the top-level entry instead of putting the list items directly under e.g. resources or images without two spaces:
kubernetes-sigs/kustomize#3946

It seems as if it is optional, I see the old more compact notation in e.g. dashboard base, working fine with kfctl used by the opendatahub-operator applying the manifests. My kustomize v4.5.6, too, has had no issue with the no-leading spaces notation as in https://github.com/opendatahub-io/odh-manifests/blob/master/odh-dashboard/base/kustomization.yaml#L25

This comment has been minimized.

Copy link
@lucferbux

lucferbux Jun 5, 2023

Author Contributor

Oh, really nice thread, I didn't think about breaking changes in marshalling when doind that kind of refactoring.
The intention behind this is to standarize the styling, I see more and more using the indent one rather than the compact, and as we are using multiple files the intention was to unify the style.
This was tested with the suggested kustomize version in the operator, so it shouldn't break anything, but thanks for pointing it out cause i wasn't aware that this change might have cause some issues.

newName: quay.io/opendatahub/openvino_model_server
digest: sha256:20dbfbaf53d1afbd47c612d953984238cb0e207972ed544a5ea662c2404f276d
60 changes: 60 additions & 0 deletions manifests/modelserving/ovms-gpu-ootb.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,60 @@
kind: Template
apiVersion: template.openshift.io/v1
metadata:
name: ovms-gpu
labels:
opendatahub.io/dashboard: 'true'
opendatahub.io/ootb: 'true'
opendatahub.io/configurable: 'true'
annotations:
tags: 'ovms,servingruntime'
description: 'OpenVino with GPU Support Model Serving Definition'
objects:
- apiVersion: serving.kserve.io/v1alpha1
kind: ServingRuntime
metadata:
name: ovms-gpu
annotations:
openshift.io/display-name: 'OpenVINO Model Server (Supports GPUs)'
labels:
opendatahub.io/dashboard: 'true'
spec:
builtInAdapter:
env:
- name: OVMS_FORCE_TARGET_DEVICE
value: NVIDIA
memBufferBytes: 134217728
modelLoadingTimeoutMillis: 90000
runtimeManagementPort: 8888
serverType: ovms
containers:
- args:
- '--port=8001'
- '--rest_port=8888'
- '--config_path=/models/model_config_list.json'
- '--file_system_poll_wait_seconds=0'
- '--grpc_bind_address=127.0.0.1'
- '--rest_bind_address=127.0.0.1'
image: ovms-1
name: ovms
resources:
limits:
cpu: '0'
memory: 0Gi
requests:
cpu: '0'
memory: 0Gi
grpcDataEndpoint: 'port:8001'
grpcEndpoint: 'port:8085'
multiModel: true
protocolVersions:
- grpc-v1
replicas: 1
supportedModelFormats:
- autoSelect: true
name: openvino_ir
version: opset1
- autoSelect: true
name: onnx
version: '1'
parameters: []
Original file line number Diff line number Diff line change
Expand Up @@ -16,6 +16,7 @@ objects:
name: ovms
annotations:
openshift.io/display-name: 'OpenVINO Model Server'
opendatahub.io/disable-gpu: 'true'

This comment has been minimized.

Copy link
@shalberd

shalberd Jun 3, 2023

Contributor

so the image is the same, just this annotation differs, correct?

This comment has been minimized.

Copy link
@andrewballantyne

andrewballantyne Jun 5, 2023

Member

@shalberd Names are slightly different for selection in dropdown:

        openshift.io/display-name: 'OpenVINO Model Server (Supports GPUs)'

vs

        openshift.io/display-name: 'OpenVINO Model Server'
labels:
opendatahub.io/dashboard: 'true'
spec:
Expand Down

0 comments on commit 5d92707

Please sign in to comment.