Updated runtime to use midstream openmpi-cuda image

Author: ChughShilpa

…rks/kftv2-mpi-ddp-sft-ddp/1.5b/README.md benchmarks/kftv2-mpi-ddp-sft/README.mdbenchmarks/kftv2-mpi-ddp-sft-ddp/1.5b/README.md renamed to benchmarks/kftv2-mpi-ddp-sft/README.md benchmarks/kftv2-mpi-ddp-sft-ddp/1.5b/README.md renamed to benchmarks/kftv2-mpi-ddp-sft/README.md

@@ -11,8 +11,8 @@ Distributed Supervised Fine-Tuning benchmark using **PyTorch DDP** with **MPI**
 | Dataset | openai/gsm8k (~7.5 K grade-school math) |
 | Comm backend | **MPI** (`torch.distributed.init_process_group(backend="mpi")`) |
 | Gradient sync | DDP automatic allreduce via MPI |
-| Runtime | `mpi-cuda-openmpi-benchmark` ClusterTrainingRuntime |
-| Image | `quay.io/ksuta/odh-mpi-cuda:0.0.14` |
+| Runtime | `openmpi-cuda-benchmark` ClusterTrainingRuntime |
+| Image | `quay.io/opendatahub/odh-training-cuda130-torch210-py312-openmpi41:odh-stable` |
 
 ### MPI communication patterns exercised
 

…v2-mpi-ddp-sft-ddp/1.5b/mpi-runtime.yaml …marks/kftv2-mpi-ddp-sft/mpi-runtime.yamlbenchmarks/kftv2-mpi-ddp-sft-ddp/1.5b/mpi-runtime.yaml renamed to benchmarks/kftv2-mpi-ddp-sft/mpi-runtime.yaml benchmarks/kftv2-mpi-ddp-sft-ddp/1.5b/mpi-runtime.yaml renamed to benchmarks/kftv2-mpi-ddp-sft/mpi-runtime.yaml

@@ -6,7 +6,7 @@
 apiVersion: trainer.kubeflow.org/v1alpha1
 kind: ClusterTrainingRuntime
 metadata:
-  name: mpi-cuda-openmpi-benchmark
+  name: openmpi-cuda-benchmark
   labels:
     trainer.kubeflow.org/framework: openmpi
 spec:
@@ -33,7 +33,7 @@ spec:
             template:
               spec:
                 containers:
-                - image: quay.io/ksuta/odh-mpi-cuda:0.0.14
+                - image: quay.io/opendatahub/odh-training-cuda130-torch210-py312-openmpi41:odh-stable
                   name: node
                   resources:
                     limits:
@@ -55,7 +55,7 @@ spec:
                   command:
                   - /usr/local/bin/uid_entrypoint.sh
                   - /usr/sbin/sshd
-                  image: quay.io/ksuta/odh-mpi-cuda:0.0.14
+                  image: quay.io/opendatahub/odh-training-cuda130-torch210-py312-openmpi41:odh-stable
                   name: node
                   readinessProbe:
                     initialDelaySeconds: 3

…v2-mpi-ddp-sft-ddp/1.5b/train_sft_ddp.py …marks/kftv2-mpi-ddp-sft/train_sft_ddp.pybenchmarks/kftv2-mpi-ddp-sft-ddp/1.5b/train_sft_ddp.py renamed to benchmarks/kftv2-mpi-ddp-sft/train_sft_ddp.py benchmarks/kftv2-mpi-ddp-sft-ddp/1.5b/train_sft_ddp.py renamed to benchmarks/kftv2-mpi-ddp-sft/train_sft_ddp.py

@@ -11,7 +11,7 @@ spec:
   runtimeRef:
     apiGroup: trainer.kubeflow.org
     kind: ClusterTrainingRuntime
-    name: mpi-cuda-openmpi-benchmark
+    name: openmpi-cuda-benchmark
   trainer:
     command:
     - /usr/local/bin/uid_entrypoint.sh