MaggotUBA breaks with GPU acceleration on

larvatagger.sh predict path/to/datafile --gpus all breaks with the following error message:

Traceback (most recent call last):
  File "/app/MaggotUBA/src/maggotuba/models/predict_model.py", line 191, in <module>
    main(predict_model)
  File "/app/MaggotUBA/.venv/lib/python3.9/site-packages/taggingbackends/main.py", line 164, in main
    return fun(backend, **args)
  File "/app/MaggotUBA/src/maggotuba/models/predict_model.py", line 52, in predict_model
    predict_individual_data_files(backend, model, input_files_and_labels)
  File "/app/MaggotUBA/src/maggotuba/models/predict_model.py", line 121, in predict_individual_data_files
    predictions = model.predict((t[larva], spines))
  File "/app/MaggotUBA/.venv/lib/python3.9/site-packages/torch/autograd/grad_mode.py", line 27, in decorate_context
    return func(*args, **kwargs)
  File "/app/MaggotUBA/src/maggotuba/models/trainers.py", line 213, in predict
    output = self.forward(data)
  File "/app/MaggotUBA/src/maggotuba/models/trainers.py", line 132, in forward
    y = self.model(x.to(self.device))
  File "/app/MaggotUBA/.venv/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl
    return forward_call(*input, **kwargs)
  File "/app/MaggotUBA/src/maggotuba/models/modules.py", line 375, in forward
    return self.clf(self.encoder(x))
  File "/app/MaggotUBA/.venv/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl
    return forward_call(*input, **kwargs)
  File "/app/MaggotUBA/src/maggotuba/models/modules.py", line 69, in forward
    return self.model(x)
  File "/app/MaggotUBA/.venv/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl
    return forward_call(*input, **kwargs)
  File "/app/MaggotUBA/.venv/lib/python3.9/site-packages/behavior_model/models/neural_nets.py", line 31, in forward
    convolutions = self.convolutional_blocks(x)
  File "/app/MaggotUBA/.venv/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl
    return forward_call(*input, **kwargs)
  File "/app/MaggotUBA/.venv/lib/python3.9/site-packages/torch/nn/modules/container.py", line 204, in forward
    input = module(input)
  File "/app/MaggotUBA/.venv/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl
    return forward_call(*input, **kwargs)
  File "/app/MaggotUBA/.venv/lib/python3.9/site-packages/torch/nn/modules/conv.py", line 463, in forward
    return self._conv_forward(input, self.weight, self.bias)
  File "/app/MaggotUBA/.venv/lib/python3.9/site-packages/torch/nn/modules/conv.py", line 459, in _conv_forward
    return F.conv2d(input, weight, bias, self.stride,
RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same

This happens with the Docker image only, but it's not clear whether GPU acceleration is enabled with larvatagger.jl predict.

EDIT: Reproduced with JULIA_PROJECT=$(pwd)/TaggingBackends LarvaTagger/scripts/larvatagger.jl predict MaggotUBA-adapter 20221005-1 path/to/datafile

This happens if NVIDIA GPUs are available on the host.