Add NM_ARCH env for deepsparse

Signed-off-by: Francesco Murdaca <fmurdaca@redhat.com>
AICoE · Sep 30, 2021 · a2cdb65 · a2cdb65
1 parent 42c8499
commit a2cdb65
Show file tree

Hide file tree

Showing 4 changed files with 25 additions and 8 deletions.
diff --git a/manifests/overlays/nm-inference/deploymentconfig.yaml b/manifests/overlays/nm-inference/deploymentconfig.yaml
@@ -28,6 +28,8 @@ spec:
               value: "1"
             - name: THOTH_AIDEVSECOPS_MODEL_VERSION
               value: "torch-210921164335-c352fe9b17e2f837_mnist_classification_pruned"
+            - name: NM_ARCH
+              value: "avx512"
           imagePullPolicy: Always
           ports:
             - containerPort: 8080

diff --git a/manifests/overlays/pytorch-inference/deploymentconfig.yaml b/manifests/overlays/pytorch-inference/deploymentconfig.yaml
@@ -24,7 +24,7 @@ spec:
           env:
             - name: WEB_CONCURRENCY
               value: "1"
-            - name: USE_PYTORCH
+            - name: TUTORIAL_USE_PYTORCH
               value: "1"
             - name: THOTH_AIDEVSECOPS_MODEL_VERSION
               value: "torch-210921163030-5341ad0f6f389a55"

diff --git a/src/pytorch_model.py b/src/pytorch_model.py
@@ -136,4 +136,4 @@ def predict(self, image):
         with torch.no_grad():
             output, last_layer = self.model(image_)
             pred_y = torch.max(output, 1)[1].data.squeeze()
-            return pred_y, output.tolist()[0][pred_y]
+            return pred_y, last_layer.tolist()[0][pred_y]
diff --git a/wsgi.py b/wsgi.py
@@ -61,20 +61,35 @@
     from src.neural_magic_model import Model as NeuralMagicModel
 
     nm_model = NeuralMagicModel()
+    # custom metric to expose model version
+    model_version_metric = prometheus_metrics.info(
+        "aidevsecops_tutorial_model_info",
+        "Model version deployed",
+        version=nm_model.model_version,  # label
+    )
+
 elif USE_PYTORCH:
     from src.pytorch_model import Model as PytorchModel
 
     pytorch_model = PytorchModel()
+    # custom metric to expose model version
+    model_version_metric = prometheus_metrics.info(
+        "aidevsecops_tutorial_model_info",
+        "Model version deployed",
+        version=pytorch_model.model_version,  # label
+    )
+
 else:
     from src.model import Model as TensorflowModel
+
     model = TensorflowModel()
 
-# custom metric to expose model version
-model_version_metric = prometheus_metrics.info(
-    "aidevsecops_tutorial_model_info",
-    "Model version deployed",
-    version=model.model_version,  # label
-)
+    # custom metric to expose model version
+    model_version_metric = prometheus_metrics.info(
+        "aidevsecops_tutorial_model_info",
+        "Model version deployed",
+        version=model.model_version,  # label
+    )
 
 
 @application.before_first_request