Mode Block is null after loading from pretrained model file?

[anasbinwaquar]

I'm trying to run a PyTorch model which I trained on Python. I'm trying to run it in java to see speed differences whether it's any faster or not. I converted the model to Torchscript via this script.

model = MyCustomModel(num_classes=101, pretrained=True)
print(model.state_dict())
model.eval()
inputs = torch.rand(1, 3, 8, 112, 112)
traced_cell = torch.jit.trace(model, inputs)
traced_cell.save('model.zip')

Now in Java when I load the model using this code Snippet it still shows the model Block is null

        Path modelDir = Paths.get("F:\\Java-Inferencing\\Model");
        Model model = Model.newInstance("model.zip");
        System.out.println(model);
        System.out.println(model.getBlock());

Does DJL not support User-defined PyTorch models?

[frankfliu]

You need load the model:

        Path modelDir = Paths.get("F:\\Java-Inferencing\\Model");
        Model model = Model.newInstance("MyJavaModel");
        model.load(modelDir);  // <---- load the model here
        System.out.println(model);
        System.out.println(model.getBlock());