RuntimeError: element 0 of tensors does not require grad and does not have a grad_fn

[sleepingcat4]

Thanks for sharing such oustanding tutorials on Transfer-learning and Quantum Circuits. I was experimenting the code with MobileNet V2. By default the notebook uses Resnet18. But, I thought to use MobileNet V2 for experimentation. Unfortunately, I am receiving this error. Can you please help me in revolving this error?

Code I changed:

model_hybrid = torchvision.models.mobilenet_v2(pretrained=True)

Error received:

Training started:
---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
[<ipython-input-21-efc42279baeb>](https://localhost:8080/#) in <cell line: 1>()
----> 1 model_hybrid = train_model(model_hybrid, criterion, optimizer_hybrid,exp_lr_scheduler, num_epochs=num_epochs)

2 frames
[/usr/local/lib/python3.10/dist-packages/torch/autograd/__init__.py](https://localhost:8080/#) in backward(tensors, grad_tensors, retain_graph, create_graph, grad_variables, inputs)
    198     # some Python versions print out the first line of a multi-line function
    199     # calls in the traceback and some print out the last line
--> 200     Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
    201         tensors, grad_tensors_, retain_graph, create_graph, inputs,
    202         allow_unreachable=True, accumulate_grad=True)  # Calls into the C++ engine to run the backward pass

RuntimeError: element 0 of tensors does not require grad and does not have a grad_fn

Notebook used: c2q_transfer_learning_ants_bees.ipynb

[KetpuntoG]

Hi! In this case, the issue is here:

model_hybrid = torchvision.models.resnet18(pretrained=True)

for param in model_hybrid.parameters():
    param.requires_grad = False
    
if quantum:
    model_hybrid.fc = Quantumnet()
    
elif classical_model == '512_2':
    model_hybrid.fc = nn.Linear(512, 2)
    
elif classical_model == '512_nq_2':
    model_hybrid.fc = nn.Sequential(nn.Linear(512, n_qubits), torch.nn.ReLU(), nn.Linear(n_qubits, 2)) 

elif classical_model == '551_512_2':
    model_hybrid.fc = nn.Sequential(nn.Linear(512, 512), torch.nn.ReLU(), nn.Linear(512, 2))

# Use CUDA or CPU according to the "device" object.
model_hybrid = model_hybrid.to(device)

This cell is built taking into account the resnet model. For example, the attribute "fc" does not exist in the model you propose. So a more detailed study would have to be done to change to another one.

It would be great to see the results with other models 😄

[CatalinaAlbornoz]

Hi @sleepingcat4, to complement @KetpuntoG's answer,

The attributes in ResNet and MobileNetV2 are different. You can check the source code to see this.
Maybe this stackoverflow post can help you. I haven't tested it myself but it seems like it worked for some people.

Let us know if this helps!

[sleepingcat4]

thank you for the comments. @CatalinaAlbornoz @KetpuntoG I was aware of the differences between architecture but I was not sure which particular code/code cell was causing the error. I am not sure, whether it was the QuantumNet circuit or when I was calling the QuantumNet on the pretained-model.

I will keep you guys updated and maybe able to share results with you as well after a few days.

[CatalinaAlbornoz]

Yes please let us know how it goes @sleepingcat4 !