def get_qconfig(per_channel):
if per_channel:
return torch.quantization.get_default_qconfig('fbgemm')
else:
act = MovingAverageMinMaxObserver.with_args(reduce_range=False)
return torch.quantization.QConfig(activation=act,
weight=default_weight_observer)
print('\n Inverted Residual Block: After fusion\n\n',
float_model.features[1].conv)
print("Size of baseline model")
original_size = print_size_of_model(float_model)
top1_fp32, top5_fp32 = eval_accuracy(float_model, data_dir)
print('\nFP32 accuracy: %2.2f' % top1_fp32.avg)
per_tensor_quantized_model = load_model(saved_model_dir +
float_model_file).to('cpu')
per_tensor_quantized_model.eval()
# Fuse Conv, bn and relu
per_tensor_quantized_model.fuse_model()
act = MovingAverageMinMaxObserver.with_args(reduce_range=False)
per_tensor_quantized_model.qconfig = torch.quantization.QConfig(
activation=act, weight=default_weight_observer)
print(per_tensor_quantized_model.qconfig)
torch.quantization.prepare(per_tensor_quantized_model, inplace=True)
# Calibrate first
print('Post Training Quantization Prepare: Inserting Observers')
print('\n Inverted Residual Block:After observer insertion \n\n',
per_tensor_quantized_model.features[1].conv)
# Calibrate with the training set
for image, _ in get_train_loader("imagenet_1k"):
per_tensor_quantized_model(image)
print('Post Training Quantization: Calibration done')