commit

lab3/lab3.py

@@ -25,39 +25,40 @@
 
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 
+# Load regular model
 
-# Create regular pytorch model
+if not trt:
 
-print("loading model...")
+    print("loading model...")
 
-timest = time.time()
-
-# model = torch.hub.load('pytorch/vision:v0.8.0', 'wide_resnet101_2', pretrained=True).eval().cuda()
-model = alexnet(pretrained = True).eval().cuda()
+    timest = time.time()
 
-print("model loaded in {}s".format(round(time.time() - timest, 3)))
+    # model = torch.hub.load('pytorch/vision:v0.8.0', 'wide_resnet101_2', pretrained=True).eval().cuda()
+    model = alexnet(pretrained = True).eval().cuda()
 
+    print("model loaded in {}s".format(round(time.time() - timest, 3)))
 
 # Load TRT
 
-if trt:
+else:
 
     print("loding trt model...")
     timesttrt = time.time()
 
-    try:
-        model = torch.load(MODEL_TRT_PATH)
+    try: # Load from file
+        model_trt = TRTModule()
+        model_trt.load_state_dict(torch.load(MODEL_TRT_PATH))
 
-    except FileNotFoundError:
+    except FileNotFoundError: # Convert from regular
 
         print("converting torch to trt...")
 
         x = torch.ones((1, 3, 224, 224)).cuda()
 
         timest = time.time()
-        model_trt = torch2trt(model, [x])
+        model_trt = torch2trt(alexnet(pretrained = True).eval().cuda(), [x])
 
-        torch.save(model_trt, MODEL_TRT_PATH)
+        torch.save(model_trt.state_dict(), MODEL_TRT_PATH)
 
         print("converted in {}s".format(round(time.time() - timest, 3)))