🔖 Adjust docker image

Author: ZihaoZhou

mnist/interactive.yaml

@@ -7,7 +7,7 @@ spec:
   - name: gpu-container
     # torch113_cuda117_ds076
     # image: docker.io/deepspeed/deepspeed:v072_torch112_cu117
-    image: docker.io/zihaokevinzhou/deepspeed:torch113_cuda117_ds076
+    image: gitlab-registry.nrp-nautilus.io/zihaozhou/nautilus_tutorial:torch113_cuda117_ds076
     imagePullPolicy: Always
     command: ["sleep", "infinity"]
     volumeMounts:
@@ -34,10 +34,6 @@ spec:
             operator: In
             values:
             - NVIDIA-GeForce-RTX-3090
-          - key: kubernetes.io/hostname
-            operator: In
-            values:
-            - k8s-3090-02.clemson.edu
   volumes:
     - name: stpp-vol
       persistentVolumeClaim:

mnist/mnist.yaml

@@ -5,13 +5,12 @@ apiVersion: batch/v1
 kind: Job
 metadata:
   name: mnist
-  namespace: deep-forecast
 spec:
   template:
     spec:
       containers:
       - name: gpu-container
-        image: docker.io/horovod/horovod:sha-811cf67
+        image: gitlab-registry.nrp-nautilus.io/zihaozhou/nautilus_tutorial:hovorod
         command: ["/bin/bash","-c"]
         # NCCL_DEBUG=INFO
         args: ["git clone https://github.com/Rose-STL-Lab/nautilus_tutorial.git;
@@ -41,10 +40,6 @@ spec:
                 operator: In
                 values:
                 - NVIDIA-GeForce-RTX-3090
-              - key: kubernetes.io/hostname
-                operator: In
-                values:
-                - k8s-3090-02.clemson.edu
       volumes:
         - name: stpp-vol
           persistentVolumeClaim:

mnist/mnist_ddp.yaml

@@ -5,13 +5,12 @@ apiVersion: batch/v1
 kind: Job
 metadata:
   name: mnist
-  namespace: deep-forecast
 spec:
   template:
     spec:
       containers:
       - name: gpu-container
-        image: horovod/horovod:sha-811cf67
+        image: gitlab-registry.nrp-nautilus.io/zihaozhou/nautilus_tutorial:hovorod
         command: ["/bin/bash","-c"]
         # NCCL_DEBUG=INFO
         args: ["git clone https://github.com/Rose-STL-Lab/nautilus_tutorial.git;
@@ -41,10 +40,6 @@ spec:
                 operator: In
                 values:
                 - NVIDIA-GeForce-RTX-3090
-              - key: kubernetes.io/hostname
-                operator: In
-                values:
-                - k8s-3090-02.clemson.edu
       volumes:
         - name: stpp-vol
           persistentVolumeClaim:

mnist/mnist_ds.yaml

@@ -5,13 +5,12 @@ apiVersion: batch/v1
 kind: Job
 metadata:
   name: mnist
-  namespace: deep-forecast
 spec:
   template:
     spec:
       containers:
       - name: gpu-container
-        image: docker.io/zihaokevinzhou/deepspeed:torch113_cuda117_ds076
+        image: gitlab-registry.nrp-nautilus.io/zihaozhou/nautilus_tutorial:torch113_cuda117_ds076
         command: ["/bin/bash","-c"]
         # NCCL_DEBUG=INFO
         args: ['git clone https://github.com/Rose-STL-Lab/nautilus_tutorial.git;
@@ -41,10 +40,6 @@ spec:
                 operator: In
                 values:
                 - NVIDIA-GeForce-RTX-3090
-              - key: kubernetes.io/hostname
-                operator: In
-                values:
-                - k8s-3090-02.clemson.edu
       volumes:
         - name: stpp-vol
           persistentVolumeClaim:

mnist/mnist_hf.yaml

@@ -5,13 +5,12 @@ apiVersion: batch/v1
 kind: Job
 metadata:
   name: mnist
-  namespace: deep-forecast
 spec:
   template:
     spec:
       containers:
       - name: gpu-container
-        image: horovod/horovod:sha-811cf67
+        image: gitlab-registry.nrp-nautilus.io/zihaozhou/nautilus_tutorial:hovorod
         command: ["/bin/bash","-c"]
         # NCCL_DEBUG=INFO
         args: ['git clone https://github.com/Rose-STL-Lab/nautilus_tutorial.git;
@@ -43,10 +42,6 @@ spec:
                 operator: In
                 values:
                 - NVIDIA-GeForce-RTX-3090
-              - key: kubernetes.io/hostname
-                operator: In
-                values:
-                - k8s-3090-02.clemson.edu
       volumes:
         - name: stpp-vol
           persistentVolumeClaim:

mnist/mnist_hvd.yaml

@@ -5,13 +5,12 @@ apiVersion: batch/v1
 kind: Job
 metadata:
   name: mnist
-  namespace: deep-forecast
 spec:
   template:
     spec:
       containers:
       - name: gpu-container
-        image: docker.io/horovod/horovod:sha-811cf67
+        image: gitlab-registry.nrp-nautilus.io/zihaozhou/nautilus_tutorial:hovorod
         command: ["/bin/bash","-c"]
         # NCCL_DEBUG=INFO
         args: ["git clone https://github.com/Rose-STL-Lab/nautilus_tutorial.git;
@@ -41,10 +40,6 @@ spec:
                 operator: In
                 values:
                 - NVIDIA-GeForce-RTX-3090
-              - key: kubernetes.io/hostname
-                operator: In
-                values:
-                - k8s-3090-02.clemson.edu
       volumes:
         - name: stpp-vol
           persistentVolumeClaim: