How to make sure trained models use GPU for inference?

[mg10011]

I am having the following problem after training my DANet model. When I reload the model after saving it using the 'load_model()' method, my GPU shows almost no usage. Thus, the loaded model is being run on the CPU. Moreover, unlike other PyTorch models, I cannot use the 'model.to(cuda_device)' as the 'to()' method is not defined. Please advise how I can insure that my loaded model will use the GPU. Thanks in advance.

[mg10011]

Additional info:

When I run my fitted model, I get the following results:

    In [11]: for j in [0, 1]:
    ...:     torch.cuda.empty_cache()
    ...:
    ...:     # Get the current GPU memory usage before inference
    ...:     gpu_memory_before = torch.cuda.memory_allocated()
    ...:
    ...:     Y_valid_pred_scaled_models[j] = tabular_model[j].predict(test=df_valid).to_numpy()
    ...:     Y_test_pred_scaled_models[j] = tabular_model[j].predict(test=df_test).to_numpy()
    ...:
    ...:     # Get the GPU memory usage after inference
    ...:     gpu_memory_after = torch.cuda.memory_allocated()
    ...:
    ...:     # Calculate the difference in GPU memory usage
    ...:     gpu_memory_used = gpu_memory_after - gpu_memory_before
    ...:
    ...:     print(f"Model {j+1}: GPU memory used during inference: {gpu_memory_used} bytes")
    ...:
Model 1: GPU memory used during inference: 0 bytes
Model 2: GPU memory used during inference: 0 bytes`

Thus, for some reason my tabular_model object is not utilizing the GPU for inference.