Executing the example leads to an error of mismatched devices. The model is initialized on the CPU, while during initialization weights are loaded on CUDA. Pytorch doesn't allow mismatched devices between tensors anymore. Solution: handle everythingo on the cpu until initialization is done, then move the model over to CUDA.
Executing the example leads to an error of mismatched devices. The model is initialized on the CPU, while during initialization weights are loaded on CUDA. Pytorch doesn't allow mismatched devices between tensors anymore. Solution: handle everythingo on the cpu until initialization is done, then move the model over to CUDA.