How can I implement pytorchjob's workers to use different images or configurations?

[certainly-cyber]

What happened?

Hello everyone! As mentioned above, I hope that the worker of pytorhjob can use different images or configurations.

Normally, my yaml is similar to this：

apiVersion: "kubeflow.org/v1"
kind: "PyTorchJob"
metadata:
name: "resnet-1"
spec:
pytorchReplicaSpecs:
Master:
replicas: 1
...
Worker:
replicas: 3
...

However, if I want to use different configurations (such as images) for different workers, how should I do it? I tried to configure multiple workers, like this：

spec:
pytorchReplicaSpecs:
Master:
replicas: 1
...
Worker:
replicas: 1
...
Worker:
replicas: 1
...

But unfortunately, this doesn't work.

What did you expect to happen?

How can I implement pytorchjob's workers to use different images or configurations?

Environment

Kubernetes version:
Client Version: v1.30.2
Kustomize Version: v5.0.4-0.20230601165947-6ce0bf390ce3
Server Version: v1.30.2

Training Operator version:
kubeflow/training-operator:v1-9e52eb7#

Impacted by this bug?

Give it a 👍 We prioritize the issues with most 👍

[tenzen-y]

Thank you for creating this issue!
Currently, the training-operator does not support the requested behavior. The PyTorchJob does not allow us to specify duplicate roles in a single job.

/remove-kind bug
/kind support

[google-oss-prow]

@tenzen-y: The label(s) kind/support cannot be applied, because the repository doesn't have them.

In response to this:

Thank you for creating this issue!
Currently, the training-operator does not support the requested behavior. The PyTorchJob does not allow us to specify duplicate roles in a single job.

/remove-kind bug
/kind support

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.

[certainly-cyber]

Thank you for creating this issue! Currently, the training-operator does not support the requested behavior. The PyTorchJob does not allow us to specify duplicate roles in a single job.

/remove-kind bug /kind support

Ok, I got it, thank you for your reply~ By the way, if I still want to achieve this behavior, is there any way to do it? I can accept any other job (tfjob or anythong else) or any proposal, thanks again and have a nice day.