Issue using Pytorch quantized models

[nik1806]

Traceback (most recent call last):
File "driver.py", line 221, in
main()
File "driver.py", line 197, in main
pilot = Pilot(app_configuration, controller, app_configuration.brain_path)
File "/root/BehaviorMetrics/behavior_metrics/pilot.py", line 69, in init
self.initialize_robot()
File "/root/BehaviorMetrics/behavior_metrics/pilot.py", line 115, in initialize_robot
self.configuration.experiment_model, self.configuration.brain_kwargs)
File "/root/BehaviorMetrics/behavior_metrics/brains/brains_handler.py", line 28, in init
self.load_brain(brain_path)
File "/root/BehaviorMetrics/behavior_metrics/brains/brains_handler.py", line 55, in load_brain
self.active_brain = Brain(self.sensors, self.actuators, model=self.model, handler=self, config=self.config)
File "/root/BehaviorMetrics/behavior_metrics/brains/f1/brain_f1_torch.py", line 64, in init
self.net = torch.jit.load(PRETRAINED_MODELS + model).to(self.device)
File "/usr/local/lib/python3.7/dist-packages/torch/jit/_serialization.py", line 161, in load
cpp_module = torch._C.import_ir_module(cu, str(f), map_location, _extra_files)
RuntimeError:
Unknown type name 'NoneType':
Serialized File "code/torch/torch/nn/quantized/modules/linear.py", line 14
return (qweight, bias, training, dtype)
def setstate(self: torch.torch.nn.quantized.modules.linear.LinearPackedParams,
state: Tuple[Tensor, Optional[Tensor], bool, int]) -> NoneType:
~~~~~~~~ <--- HERE
self.dtype = (state)[3]
_0 = (self).set_weight_bias((state)[0], (state)[1], )

[nik1806]

The problem was with installed Pytorch version. I upgraded it to 1.11.0 and everything worked as expected.

[sergiopaniego]

We should update the requirements.txt then, right? @nik1806

[nik1806]

Yes, we can update to latest work of PyTorch.

[sergiopaniego]

Great! Can you open the new PR, please?