Merge pull request #4413 from harshthakkar01/image-build

Fix nvidia version mismatch for service images

Author: harshthakkar01

examples/machine-learning/build-service-images/common/blueprint.yaml

@@ -82,6 +82,24 @@ deployment_groups:
               ansible.builtin.apt:
                 deb: "{{ cuda_repo_filename }}"
                 state: present
+            # The following 2 tasks work around a temporary issue with Ubuntu
+            # packaging of NVIDIA 570 driver series for kernel 6.8.0-1032
+            # This command ensures that any holds are removed before attempting an upgrade.
+            # We ignore failures in case the packages were not held.
+            - name: Unhold NVIDIA driver packages
+              ansible.builtin.command:
+                cmd: apt-mark unhold linux-modules-nvidia-570-server-open-gcp linux-modules-nvidia-570-server-open-6.8.0-1032-gcp
+              become: true
+              changed_when: false
+              failed_when: false
+            - name: Install latest NVIDIA driver metapackage and kernel module
+              ansible.builtin.apt:
+                name:
+                  - linux-modules-nvidia-570-server-open-gcp
+                  - linux-modules-nvidia-570-server-open-6.8.0-1032-gcp
+                state: latest
+                update_cache: yes
+              become: true
             - name: Reduce NVIDIA repository priority
               ansible.builtin.copy:
                 dest: /etc/apt/preferences.d/cuda-repository-pin-600