Update testing to verify the infer result

Author: krishung5

qa/L0_libtorch_instance_group_kind_model/client.py

@@ -74,23 +74,19 @@ def test_infer(self):
         outputs.append(
             httpclient.InferRequestedOutput('OUTPUT__1', binary_data=True))
 
-        if model_name == "libtorch_instance_kind_err":
-            with self.assertRaises(InferenceServerException) as ex:
-                triton_client.infer(model_name, inputs, outputs=outputs)
-            self.assertIn(
-                "Expected all tensors to be on the same device, but found at least two devices",
-                str(ex.exception))
-            return
-
         results = triton_client.infer(model_name, inputs, outputs=outputs)
 
         output0_data = results.as_numpy('OUTPUT__0')
         output1_data = results.as_numpy('OUTPUT__1')
 
-        # Only validate the shape, as the output will differ every time the
-        # model is compiled and used on different devices.
-        self.assertEqual(output0_data.shape, (1, 4))
-        self.assertEqual(output1_data.shape, (1, 4))
+        expected_output_0 = input0_data + input0_data
+        expected_output_1 = input1_data + input1_data
+
+        self.assertEqual(output0_data.shape, (1, 16))
+        self.assertEqual(output1_data.shape, (1, 16))
+
+        self.assertTrue(np.all(expected_output_0 == output0_data))
+        self.assertTrue(np.all(expected_output_1 == output1_data))
 
 
 if __name__ == '__main__':

qa/L0_libtorch_instance_group_kind_model/gen_models.py

@@ -29,23 +29,32 @@
 import torch.nn as nn
 
 
+class SumModule(nn.Module):
+
+    def __init__(self):
+        super(SumModule, self).__init__()
+
+    def forward(self, x):
+        return torch.sum(x, dim=1)
+
+
 class TestModel(nn.Module):
 
     def __init__(self, device1, device2):
         super(TestModel, self).__init__()
         self.device1 = device1
         self.device2 = device2
-        self.layers1 = nn.Sequential(nn.Linear(16, 4),).to(self.device1)
-        self.layers2 = nn.Sequential(nn.Linear(16, 4)).to(self.device2)
+        self.layers1 = SumModule().to(self.device1)
+        self.layers2 = SumModule().to(self.device2)
 
     def forward(self, INPUT0, INPUT1):
         INPUT0 = INPUT0.to(self.device1)
         INPUT1 = INPUT1.to(self.device2)
         print('INPUT0 device: {}, INPUT1 device: {}\n'.format(
             INPUT0.device, INPUT1.device))
 
-        op0 = self.layers1(INPUT0 + INPUT0)
-        op1 = self.layers2(INPUT1 + INPUT1)
+        op0 = self.layers1(torch.stack([INPUT0, INPUT0], dim=1))
+        op1 = self.layers2(torch.stack([INPUT1, INPUT1], dim=1))
         return op0, op1
 
 

qa/L0_libtorch_instance_group_kind_model/models/libtorch_multi_devices/1/model.pt

@@ -57,6 +57,7 @@ RET=0
 
 rm -f *.log *.txt
 
+mkdir -p models/libtorch_multi_devices/1
 mkdir -p models/libtorch_multi_gpu/1
 cp models/libtorch_multi_devices/config.pbtxt models/libtorch_multi_gpu/.
 (cd models/libtorch_multi_gpu && \
@@ -72,14 +73,6 @@ if [ $? -ne 0 ]; then
 fi
 set -e
 
-# Create the model that does not set instance_group_kind to 'KIND_MODEL'
-mkdir -p models/libtorch_instance_kind_err/1
-cp models/libtorch_multi_devices/config.pbtxt models/libtorch_instance_kind_err/.
-cp models/libtorch_multi_devices/1/model.pt models/libtorch_instance_kind_err/1/.
-(cd models/libtorch_instance_kind_err && \
-    sed -i "s/name: \"libtorch_multi_devices\"/name: \"libtorch_instance_kind_err\"/" config.pbtxt && \
-    sed -i "s/kind: KIND_MODEL/kind: KIND_GPU/" config.pbtxt)
-
 run_server
 if [ "$SERVER_PID" == "0" ]; then
     echo -e "\n***\n*** Failed to start $SERVER\n***"
@@ -135,22 +128,6 @@ else
     RET=1
 fi
 
-MESSAGE="INPUT0 device: cuda:2, INPUT1 device: cuda:0"
-export MODEL_NAME='libtorch_instance_kind_err'
-python3 $CLIENT_PY >> $CLIENT_LOG 2>&1 
-if [ $? -ne 0 ]; then
-    echo -e "\n***\n*** Model $MODEL_NAME FAILED. \n***"
-    cat $CLIENT_LOG
-    RET=1
-else
-    check_test_results $TEST_RESULT_FILE $EXPECTED_NUM_TESTS
-    if [ $? -ne 0 ]; then
-        cat $CLIENT_LOG
-        echo -e "\n***\n*** Test Result Verification Failed\n***"
-        RET=1
-    fi
-fi
-
 set -e
 
 kill $SERVER_PID