Skip to content

Update RHOAI manifests to include universal workbench image configura… #26

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
kramaranya wants to merge 5 commits into from
Closed

Update RHOAI manifests to include universal workbench image configura… #26

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
d8ecc0a
Update RHOAI manifests to include universal workbench image configura…
kramaranya Nov 20, 2025
f8b7c92
Update quay repo
kramaranya Nov 21, 2025
e657541
Add ROCm and CPU ImageStreams
kramaranya Dec 3, 2025
9bbceeb
Fix syntax error in CUDA Image Stream
kramaranya Dec 3, 2025
54f0089
manifests: refactor params.env to handle multiple versions
kramaranya Dec 5, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
42 changes: 42 additions & 0 deletions manifests/rhoai/kustomization.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,6 +26,45 @@ replacements:
name: kubeflow-trainer-controller-manager
fieldPaths:
- spec.template.spec.containers.0.image
- source:
kind: ConfigMap
name: rhoai-config
version: v1
fieldPath: data.odh-kubeflow-trainer-universal-workbench-image-cuda-2025-1
targets:
- select:
group: image.openshift.io
version: v1
kind: ImageStream
name: training-hub-universal-cuda
fieldPaths:
- spec.tags.0.from.name
- source:
kind: ConfigMap
name: rhoai-config
version: v1
fieldPath: data.odh-kubeflow-trainer-universal-workbench-image-rocm-2025-1
targets:
- select:
group: image.openshift.io
version: v1
kind: ImageStream
name: training-hub-universal-rocm
fieldPaths:
- spec.tags.0.from.name
- source:
kind: ConfigMap
name: rhoai-config
version: v1
fieldPath: data.odh-kubeflow-trainer-universal-workbench-image-cpu-2025-1
targets:
- select:
group: image.openshift.io
version: v1
kind: ImageStream
name: training-hub-universal-cpu
fieldPaths:
- spec.tags.0.from.name

# Labels to add to all resources and selectors.
labels:
Expand All @@ -41,6 +80,9 @@ resources:
- ../base/webhook
# - ../third-party/jobset #uncomment if jobset should be bundled under kubeflow trainer controller manager
- runtimes
- training-hub-universal-cuda-imagestream.yaml
- training-hub-universal-rocm-imagestream.yaml
- training-hub-universal-cpu-imagestream.yaml
- kubeflow-training-roles.yaml
- monitor.yaml

Expand Down
3 changes: 3 additions & 0 deletions manifests/rhoai/params.env
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1 +1,4 @@
odh-kubeflow-trainer-controller-image=quay.io/opendatahub/trainer:v2.1.0
odh-kubeflow-trainer-universal-workbench-image-cuda-2025-1=quay.io/opendatahub/odh-training-th03-cuda128-torch28-py312-rhel9:latest
odh-kubeflow-trainer-universal-workbench-image-rocm-2025-1=quay.io/mstoklus/workbench-images:py312-rocm64-torch280-3
odh-kubeflow-trainer-universal-workbench-image-cpu-2025-1=quay.io/opendatahub/odh-training-th03-cuda128-torch28-py312-rhel9:latest-cpu
43 changes: 43 additions & 0 deletions manifests/rhoai/training-hub-universal-cpu-imagestream.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,43 @@
apiVersion: image.openshift.io/v1
kind: ImageStream
metadata:
name: training-hub-universal-cpu
labels:
opendatahub.io/notebook-image: "true"
annotations:
opendatahub.io/notebook-image-url: "https://github.com/opendatahub-io/distributed-workloads/tree/main/images/universal/training"
opendatahub.io/notebook-image-name: "Training | Jupyter | PyTorch | CPU | Python"
opendatahub.io/notebook-image-desc: "Training runtime image for CPU-based fine-tuning and distributed training."
opendatahub.io/notebook-image-order: "3"
spec:
lookupPolicy:
local: true
tags:
- name: "2025.1"
annotations:
opendatahub.io/notebook-software: |
[
{"name": "Python", "version": "v3.12"},
{"name": "PyTorch", "version": "2.9.0"},
{"name": "Kubeflow SDK", "version": "0.2.0"},
{"name": "Training Hub", "version": "v0.4.0"}
]
opendatahub.io/notebook-python-dependencies: |
[
{"name": "jupyterlab", "version": "4.4.9"},
{"name": "transformers", "version": "4.57.1"},
{"name": "accelerate", "version": "1.10.0"},
{"name": "peft", "version": "0.17.0"},
{"name": "trl", "version": "0.21.0"},
{"name": "numpy", "version": "1.26.4"},
{"name": "pandas", "version": "2.3.3"},
{"name": "matplotlib-inline", "version": "0.1.7"},
{"name": "tensorboard", "version": "2.19.0"},
{"name": "instructlab-training", "version": "0.12.1"}
]
openshift.io/imported-from: quay.io/opendatahub/odh-training-th03-cuda128-torch28-py312-rhel9
from:
kind: DockerImage
name: odh-kubeflow-trainer-universal-workbench-image-cpu-2025-1
referencePolicy:
type: Source
47 changes: 47 additions & 0 deletions manifests/rhoai/training-hub-universal-cuda-imagestream.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,47 @@
apiVersion: image.openshift.io/v1
kind: ImageStream
metadata:
name: training-hub-universal-cuda
labels:
opendatahub.io/notebook-image: "true"
annotations:
opendatahub.io/notebook-image-url: "https://github.com/opendatahub-io/distributed-workloads/tree/main/images/universal/training"
opendatahub.io/notebook-image-name: "Training | Jupyter | PyTorch | CUDA | Python"
opendatahub.io/notebook-image-desc: "Training runtime image for CUDA-based fine-tuning and distributed training."
opendatahub.io/notebook-image-order: "1"
opendatahub.io/recommended-accelerators: '["nvidia.com/gpu"]'
spec:
lookupPolicy:
local: true
tags:
- name: "2025.1"
annotations:
opendatahub.io/notebook-software: |
[
{"name": "CUDA", "version": "12.8"},
{"name": "Python", "version": "v3.12"},
{"name": "PyTorch", "version": "2.8.0"},
{"name": "Kubeflow SDK", "version": "0.2.0"},
{"name": "Training Hub", "version": "v0.4.0"}
]
opendatahub.io/notebook-python-dependencies: |
[
{"name": "jupyterlab", "version": "4.4.4"},
{"name": "flash-attn", "version": "2.8.3"},
{"name": "transformers", "version": "4.57.1"},
{"name": "accelerate", "version": "1.10.0"},
{"name": "liger-kernel", "version": "0.6.2"},
{"name": "peft", "version": "0.17.0"},
{"name": "triton", "version": "3.4.0"},
{"name": "trl", "version": "0.21.0"},
{"name": "numpy", "version": "2.3.5"},
{"name": "pandas", "version": "2.3.3"},
{"name": "matplotlib", "version": "3.10.7"},
{"name": "scikit-learn", "version": "1.7.2"}
]
openshift.io/imported-from: quay.io/opendatahub/odh-training-th03-cuda128-torch28-py312-rhel9
Copy link
Copy Markdown

@MStokluska MStokluska Nov 21, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I wonder if this should be dynamically fetched? I guess at some point this will be coming from registry (??)

Copy link
Copy Markdown
Author

@kramaranya kramaranya Dec 3, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Notebooks repo keeps it static, but we can make it dynamic after images move to the final registry

from:
kind: DockerImage
name: odh-kubeflow-trainer-universal-workbench-image-cuda-2025-1
referencePolicy:
type: Source
43 changes: 43 additions & 0 deletions manifests/rhoai/training-hub-universal-rocm-imagestream.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,43 @@
apiVersion: image.openshift.io/v1
kind: ImageStream
metadata:
name: training-hub-universal-rocm
labels:
opendatahub.io/notebook-image: "true"
annotations:
opendatahub.io/notebook-image-url: "https://github.com/opendatahub-io/distributed-workloads/tree/main/images/universal/training"
opendatahub.io/notebook-image-name: "Training | Jupyter | PyTorch | ROCm | Python"
opendatahub.io/notebook-image-desc: "Training runtime image for ROCm-based fine-tuning and distributed training."
opendatahub.io/notebook-image-order: "2"
opendatahub.io/recommended-accelerators: '["amd.com/gpu"]'
spec:
lookupPolicy:
local: true
tags:
- name: "2025.1"
annotations:
opendatahub.io/notebook-software: |
[
{"name": "ROCm", "version": "6.4"},
{"name": "Python", "version": "v3.12"},
{"name": "PyTorch", "version": "2.8.0"},
{"name": "Kubeflow SDK", "version": "0.2.0"}
]
opendatahub.io/notebook-python-dependencies: |
[
{"name": "jupyterlab", "version": "4.4.9"},
{"name": "transformers", "version": "4.55.2"},
{"name": "accelerate", "version": "1.10.0"},
{"name": "liger-kernel", "version": "0.5.10"},
{"name": "peft", "version": "0.17.0"},
{"name": "trl", "version": "0.21.0"},
{"name": "deepspeed", "version": "0.14.3"},
{"name": "datasets", "version": "4.0.0"},
{"name": "tensorboard", "version": "2.19.0"}
]
openshift.io/imported-from: quay.io/mstoklus/workbench-images
from:
kind: DockerImage
name: odh-kubeflow-trainer-universal-workbench-image-rocm-2025-1
referencePolicy:
type: Source
Loading