[DNM] Add pre_tasks to explicitly install Galaxy role requirements

gpu-validation/defaults/main.yaml

@@ -45,32 +45,36 @@
   10de:27b8: 1
 
 gpu_validation_model_tests_enabled: true  # Can be disabled for lighter CUDA-only sanity testing
-gpu_validation_model_name: TinyLlama/TinyLlama-1.1B-Chat-v1.0  # RedHatAI/Llama-3.2-1B-Instruct-FP8 for RHAIIS
+gpu_validation_model_name: RedHatAI/Llama-3.2-1B-Instruct-FP8
 # [string] (optional) HuggingFace token if required for download
 gpu_validation_model_download_hf_token:
+# [string] (optional) URL to download pre-packaged model from RH intranet (e.g., http://file-server/models/tinyllama.tar.gz)
+gpu_validation_model_intranet_url:
+# [string] (optional) Local path where pre-downloaded model is located (overrides HuggingFace download)
+gpu_validation_model_local_path:
 
 # [float] Performance threshholds
 gpu_validation_model_perf_max_avg_time_per_tok: !!float "0.03"
 gpu_validation_model_perf_max_avg_time_to_first_tok: !!float "0.3"
 
 # [string] Container image to use for model serving
-gpu_validation_workload_container_image: "docker.io/vllm/vllm-openai:latest"  # "registry.redhat.io/rhaiis/vllm-cuda-rhel9:3.0.0" for RHAIIS
+gpu_validation_workload_container_image: "registry.redhat.io/rhaiis/vllm-cuda-rhel9:3.0.0"
 # [string] (optional) registry username if required for container image download
 gpu_validation_workload_registry_username:
 # [string] (optional) registry password if required for container image download
 gpu_validation_workload_registry_password:
 # [string] Cache directory path for model downloads
 gpu_validation_workload_cache_dir: "/home/cloud-user/.cache/workload"
 # [string] Cache mount path inside container
-gpu_validation_workload_cache_mount_path: "/root/.cache/huggingface"  # "/opt/app-root/src/.cache" for RHAIIS
+gpu_validation_workload_cache_mount_path:  "/opt/app-root/src/.cache"
 # [string] Container Device options
 gpu_validation_workload_device_opts: "--device nvidia.com/gpu=all"
 # [string] Container security options
 gpu_validation_workload_security_opts: "--security-opt=label=disable"
 # [string] Container options
 gpu_validation_workload_additional_opts: "--rm -it"
 # [string] Container user namespace option
-gpu_validation_workload_userns:  # "--userns=keep-id:uid=1001" for RHAIIS
+gpu_validation_workload_userns: --userns=keep-id:uid=1001
 # [string] Shared memory size
 gpu_validation_workload_shm_size: "--shm-size=4g"
 # [string] Additional environment variables

main.yaml

@@ -2,6 +2,12 @@
 - name: GPU Validation VM setup
   hosts: control_node
   gather_facts: false
+  pre_tasks:
+    - name: Install Ansible Galaxy role requirements
+      ansible.builtin.command:
+        cmd: ansible-galaxy role install -r requirements.yaml --force
+        chdir: "{{ playbook_dir }}"
+      changed_when: true
   tasks:
     - name: VM Pre-tasks
       ansible.builtin.include_role: