❓ Questions / Help / Support

[kuruhuru]

❓ Questions and Help

Hello!
I got strange error. Please, advice any fix

I have a model that works and successfully learns. No errors.

But after I run your code:

USE_ONNX = False # change this to True if you want to test onnx model
if USE_ONNX:
!pip install -q onnxruntime

vad_model, ut = torch.hub.load(repo_or_dir='snakers4/silero-vad',
model='silero_vad',
force_reload=True,
onnx=USE_ONNX)

My model starts to generate errors when learning.

110
111 model.zero_grad()
--> 112 loss.backward()
113
114 optimizer.step()

/usr/local/lib/python3.7/dist-packages/torch/_tensor.py in backward(self, gradient, retain_graph, create_graph, inputs)
361 create_graph=create_graph,
362 inputs=inputs)
--> 363 torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs)
364
365 def register_hook(self, hook):

/usr/local/lib/python3.7/dist-packages/torch/autograd/init.py in backward(tensors, grad_tensors, retain_graph, create_graph, grad_variables, inputs)
173 Variable.execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
174 tensors, grad_tensors, retain_graph, create_graph, inputs,
--> 175 allow_unreachable=True, accumulate_grad=True) # Calls into the C++ engine to run the backward pass
176
177 def grad(

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

[snakers4]

Hi,

Our models are not designed for learning.
Also please note that somewhere in our code we most likely disable the gradients during a forward pass.

As for testing how our models efffect the learning of other models - this is out of scope of this repository.
You may want to look through / ask around the discussions, I remember someone using the model as a feature extractor.

[kuruhuru]

Thank you for the answer.

torch.set_grad_enabled(True)

This works ok after installation.

[snakers4]

Then it is most likely the disabled gradients
If you are using the models are a feature extractor, make sure not to use our convenience functions, because they may also contain code like

with torch.no_grad()
...

or similar