attribute workflows to correct runners

.github/workflows/test_api_cuda.yaml

@@ -19,7 +19,7 @@ jobs:
         image:
           [
             { torch_cuda: cu118, torch_pre_release: 0, cuda_version: 11.8.0 },
-            { torch_cuda: cu121, torch_pre_release: 1, cuda_version: 12.1.0 },
+            { torch_cuda: cu121, torch_pre_release: 1, cuda_version: 12.1.1 },
           ]
 
     runs-on: [multi-gpu, nvidia-gpu, 4-a10, ci]

.github/workflows/test_cli_cuda_onnxruntime.yaml

@@ -13,7 +13,8 @@ concurrency:
 
 jobs:
   build_image_and_run_cli_cuda_onnxruntime_tests:
-    runs-on: nvidia-gpu
+    runs-on: [single-gpu, nvidia-gpu, a10, ci]
+
     steps:
       - name: Checkout
         uses: actions/checkout@v3
@@ -28,19 +29,14 @@ jobs:
           --tag opt-bench-cuda:11.8.0
           .
 
-      - name: Get GPUs with most free memory
-        id: get_devices
-        run: |
-          echo "::set-output name=devices::$(nvidia-smi --query-gpu=memory.free,index --format=csv,noheader,nounits | sort -n -k1 | tail -n 2 | awk -F', ' '{print $2}' | xargs echo -n | sed 's/ /,/g' | awk '{print $0}')"
-
       - name: Run tests
         run: docker run
           --rm
+          --gpus all
           --pid host
           --shm-size 64G
           --env USE_CUDA="1"
           --entrypoint /bin/bash
-          --gpus '"device=${{ steps.get_devices.outputs.devices }}"'
           --volume $(pwd):/workspace/optimum-benchmark
           --workdir /workspace/optimum-benchmark
           opt-bench-cuda:11.8.0

.github/workflows/test_cli_cuda_pytorch.yaml

@@ -22,7 +22,8 @@ jobs:
             { torch_cuda: cu121, torch_pre_release: 1, cuda_version: 12.1.1 },
           ]
 
-    runs-on: nvidia-gpu
+    runs-on: [multi-gpu, nvidia-gpu, 4-a10, ci]
+
     steps:
       - name: Checkout
         uses: actions/checkout@v3
@@ -37,18 +38,13 @@ jobs:
           --tag opt-bench-cuda:${{ matrix.image.cuda_version }}
           .
 
-      - name: Get GPUs with most free memory
-        id: get_devices
-        run: |
-          echo "::set-output name=devices::$(nvidia-smi --query-gpu=memory.free,index --format=csv,noheader,nounits | sort -n -k1 | tail -n 2 | awk -F', ' '{print $2}' | xargs echo -n | sed 's/ /,/g' | awk '{print $0}')"
-
       - name: Run tests
         run: docker run
           --rm
+          --gpus all
           --pid host
           --shm-size 64G
           --env USE_CUDA="1"
-          --gpus '"device=${{ steps.get_devices.outputs.devices }}"'
           --volume $(pwd):/workspace/optimum-benchmark
           --workdir /workspace/optimum-benchmark
           --entrypoint /bin/bash

.github/workflows/test_cli_cuda_torch_ort.yaml

@@ -13,7 +13,8 @@ concurrency:
 
 jobs:
   build_image_and_run_cli_cuda_torch_ort_tests:
-    runs-on: nvidia-gpu
+    runs-on: [multi-gpu, nvidia-gpu, 4-a10, ci]
+
     steps:
       - name: Checkout
         uses: actions/checkout@v3
@@ -28,19 +29,14 @@ jobs:
           --tag opt-bench-cuda:11.8.0
           .
 
-      - name: Get GPUs with most free memory
-        id: get_devices
-        run: |
-          echo "::set-output name=devices::$(nvidia-smi --query-gpu=memory.free,index --format=csv,noheader,nounits | sort -n -k1 | tail -n 2 | awk -F', ' '{print $2}' | xargs echo -n | sed 's/ /,/g' | awk '{print $0}')"
-
       - name: Run tests
         run: docker run
           --rm
+          --gpus all
           --pid host
           --shm-size 64G
           --env USE_CUDA="1"
           --entrypoint /bin/bash
-          --gpus '"device=${{ steps.get_devices.outputs.devices }}"'
           --volume $(pwd):/workspace/optimum-benchmark
           --workdir /workspace/optimum-benchmark
           opt-bench-cuda:11.8.0

.github/workflows/test_cli_tensorrt_llm.yaml

@@ -13,7 +13,8 @@ concurrency:
 
 jobs:
   pull_image_and_run_cli_tensorrt_llm_tests:
-    runs-on: nvidia-gpu
+    runs-on: [single-gpu, nvidia-gpu, a10, ci]
+
     steps:
       - name: Checkout
         uses: actions/checkout@v3
@@ -26,18 +27,13 @@ jobs:
           --tag opt-bench-tensorrt-llm:latest
           .
 
-      - name: Get GPUs with most free memory
-        id: get_devices
-        run: |
-          echo "::set-output name=devices::$(nvidia-smi --query-gpu=memory.free,index --format=csv,noheader,nounits | sort -n -k1 | tail -n 2 | awk -F', ' '{print $2}' | xargs echo -n | sed 's/ /,/g' | awk '{print $0}')"
-
       - name: Run tests
         run: docker run
           --rm
+          --gpus all
           --pid host
           --shm-size 64G
           --env USE_CUDA="1"
-          --gpus '"device=${{ steps.get_devices.outputs.devices }}"'
           --volume $(pwd):/workspace/optimum-benchmark
           --workdir /workspace/optimum-benchmark
           --entrypoint /bin/bash

.github/workflows/test_cli_tensorrt_onnxruntime.yaml

@@ -13,7 +13,8 @@ concurrency:
 
 jobs:
   build_image_and_run_cli_tensorrt_onnxruntime_tests:
-    runs-on: nvidia-gpu
+    runs-on: [single-gpu, nvidia-gpu, a10, ci]
+
     steps:
       - name: Checkout
         uses: actions/checkout@v3
@@ -26,18 +27,13 @@ jobs:
           --tag opt-bench-tensorrt:latest
           .
 
-      - name: Get GPUs with most free memory
-        id: get_devices
-        run: |
-          echo "::set-output name=devices::$(nvidia-smi --query-gpu=memory.free,index --format=csv,noheader,nounits | sort -n -k1 | tail -n 2 | awk -F', ' '{print $2}' | xargs echo -n | sed 's/ /,/g' | awk '{print $0}')"
-
       - name: Run tests
         run: docker run
           --rm
+          --gpus all
           --pid host
           --shm-size 64G
           --env USE_CUDA="1"
-          --gpus '"device=${{ steps.get_devices.outputs.devices }}"'
           --volume $(pwd):/workspace/optimum-benchmark
           --workdir /workspace/optimum-benchmark
           --entrypoint /bin/bash