mergerd

Author: Egor-Krivov

.github/workflows/execute-test-script.yml

@@ -68,8 +68,12 @@ jobs:
                         echo 'runner_labels=["self-hosted", "amd32c", "aws"]' >> $GITHUB_OUTPUT;;
                     amd64c)
                         echo 'runner_labels=["self-hosted", "amd64c", "aws"]' >> $GITHUB_OUTPUT;;
-                    nvidia)
-                        echo 'runner_labels=["self-hosted", "nvidia", "aws"]' >> $GITHUB_OUTPUT;;
+                    genoa60c)
+                        echo 'runner_labels=["self-hosted", "genoa", "60c", "gcp"]' >> $GITHUB_OUTPUT;;
+                    a100)
+                        echo 'runner_labels=["self-hosted", "nvidia", "a100"]' >> $GITHUB_OUTPUT;;
+                    v100)
+                        echo 'runner_labels=["self-hosted", "nvidia", "v100"]' >> $GITHUB_OUTPUT;;
                     *)
                         echo "runner_labels=unknown" >> $GITHUB_OUTPUT;;
                 esac

.github/workflows/test_amd.yml

@@ -10,11 +10,12 @@ on:
             runner_type:
                 description: Type of runner to use
                 required: true
-                default: amd64c
+                default: genoa60c
                 type: choice
                 options:
                     - amd32c
                     - amd64c
+                    - genoa60c
             shutdown_cloud_runner:
                 description: Shutdown AWS cluster
                 default: true

.github/workflows/test_nvidia.yml

@@ -38,13 +38,13 @@ jobs:
             compiler: ${{ matrix.type.compiler }}
             device: ${{ matrix.type.device }}
             tag: ${{ inputs.tag }}
-            runner_type: nvidia
+            runner_type: a100
             test_script: ${{ matrix.test_script }}
         secrets:
             DB_URL: ${{ secrets.DB_URL }}
 
     shutdown:
-        runs-on: [self-hosted, aws, nvidia]
+        runs-on: a100
         if: inputs.shutdown_cloud_runner
         needs: mlp_test
         steps:

dl_bench/cnn.py

@@ -34,12 +34,14 @@ def __init__(self, params) -> None:
 
         in_shape = (3, 224, 224)
         min_batches = 10
-        DATASET_SIZE = max(10_240, batch_size * min_batches)
+        min_seconds = 20
+        warmup = 10
+        DATASET_SIZE = max(10_240, batch_size * (min_batches + warmup))
         dataset = RandomInfDataset(DATASET_SIZE, in_shape)
 
         name = params.get("name", "resnet50")
         net = get_cnn(name=name)
 
         super().__init__(
-            net=net, in_shape=in_shape, dataset=dataset, batch_size=batch_size
+            net=net, in_shape=in_shape, dataset=dataset, batch_size=batch_size, min_batches=min_batches, warmup_batches=warmup, min_seconds=min_seconds
         )

dl_bench/mlp.py

@@ -81,8 +81,10 @@ def __init__(self, params) -> None:
 
         batch_size = int(params.get("batch_size", 1024))
 
-        min_batches = 20
-        DATASET_SIZE = max(102_400, batch_size * min_batches)
+        min_batches = int(params.get("min_batches", 20))
+        min_seconds = int(params.get("min_seconds", 20))
+        warmup = 10
+        DATASET_SIZE = max(102_400, batch_size * (min_batches + warmup))
         dataset = RandomInfDataset(DATASET_SIZE, in_shape)
 
         name = params.get("name", "size5")
@@ -91,6 +93,11 @@ def __init__(self, params) -> None:
         min_seconds = int(params.get("min_seconds", 10))
 
         super().__init__(
-            net=net, in_shape=in_shape, dataset=dataset, batch_size=batch_size,\
-                min_batches=min_batches, min_seconds=min_seconds, warmup_batches=10
+            net=net,
+            in_shape=in_shape,
+            dataset=dataset,
+            batch_size=batch_size,
+            min_batches=min_batches,
+            min_seconds=min_seconds,
+            warmup_batches=warmup,
         )