Add cuda check for fp16 tests

tests/ptlflow/models/test_models.py

@@ -100,35 +100,36 @@ def test_forward() -> None:
 
 
 def test_forward_fp16() -> None:
-    model_names = ptlflow.models_dict.keys()
-    for mname in model_names:
-        if mname in EXCLUDE_MODELS_FP16:
-            continue
-
-        print(mname)
-        model_ref = ptlflow.get_model_reference(mname)
-        parser = model_ref.add_model_specific_args()
-        args = parser.parse_args([])
-
-        if mname in MODEL_ARGS_FP16:
-            for name, val in MODEL_ARGS_FP16[mname].items():
-                setattr(args, name, val)
-
-        model = ptlflow.get_model(mname, args=args)
-        model = model.eval()
-        model = model.half()
-
-        s = make_divisible(256, model.output_stride)
-        num_images = 2
-        if mname in ["videoflow_bof", "videoflow_mof"]:
-            num_images = 3
-        inputs = {"images": torch.rand(1, num_images, 3, s, s)}
-
-        if torch.cuda.is_available():
-            model = model.cuda()
-            inputs["images"] = inputs["images"].cuda().half()
-
-        model(inputs)
+    if torch.cuda.is_available():
+        model_names = ptlflow.models_dict.keys()
+        for mname in model_names:
+            if mname in EXCLUDE_MODELS_FP16:
+                continue
+
+            print(mname)
+            model_ref = ptlflow.get_model_reference(mname)
+            parser = model_ref.add_model_specific_args()
+            args = parser.parse_args([])
+
+            if mname in MODEL_ARGS_FP16:
+                for name, val in MODEL_ARGS_FP16[mname].items():
+                    setattr(args, name, val)
+
+            model = ptlflow.get_model(mname, args=args)
+            model = model.eval()
+            model = model.half()
+
+            s = make_divisible(256, model.output_stride)
+            num_images = 2
+            if mname in ["videoflow_bof", "videoflow_mof"]:
+                num_images = 3
+            inputs = {"images": torch.rand(1, num_images, 3, s, s)}
+
+            if torch.cuda.is_available():
+                model = model.cuda()
+                inputs["images"] = inputs["images"].cuda().half()
+
+            model(inputs)
 
 
 @pytest.mark.skip(