try

Author: ydshieh

.github/workflows/model_jobs.yml

@@ -17,7 +17,7 @@ on:
         type: string
       runner_map:
         required: false
-        default: '{"models/vit": ["aws-g4dn-4xlarge-cache", "aws-g4dn-12xlarge-cache"], "models/clip": ["aws-g5-4xlarge-cache", "aws-g5-12xlarge-cache"]}'
+        default: '{"models/vit": {"single-gpu": "aws-g4dn-4xlarge-cache", "multi-gpu": "aws-g4dn-12xlarge-cache"}, "models/clip": {"single-gpu": "aws-g5-4xlarge-cache", "multi-gpu": "aws-g5-12xlarge-cache"}}'
         type: string
       docker:
         required: true
@@ -47,10 +47,9 @@ jobs:
       max-parallel: 8
       fail-fast: false
       matrix:
-        machine_type: ${{ fromJson(inputs.runner_map)[matrix.folders] }}
         folders: ${{ fromJson(inputs.folder_slices)[inputs.slice_id] }}
     runs-on:
-      group: '${{ inputs.machine_type }}'
+      group: ${{ fromJson(inputs.runner_map)[matrix.folders][inputs.machine_type] }}
     container:
       image: ${{ inputs.docker }}
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/

.github/workflows/self-scheduled.yml

@@ -111,11 +111,12 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
+        machine_type: [single-gpu, multi-gpu]
         slice_id: ${{ fromJSON(needs.setup.outputs.slice_ids) }}
     uses: ./.github/workflows/model_jobs.yml
     with:
       folder_slices: ${{ needs.setup.outputs.folder_slices }}
-      machine_type: ''
+      machine_type: ${{ matrix.machine_type }}
       slice_id: ${{ matrix.slice_id }}
       runner: ${{ inputs.runner }}
       docker: ${{ inputs.docker }}