brevitas
brevitas copied to clipboard
Error when using IntQuant
When creating my own integer quantizer, I get the following error:
TypeError: forward() missing 3 required positional arguments: 'zero_point', 'bit_width', and 'x'
Using this quantizer:
class ActInt(ExtendedInjector):
tensor_quant = IntQuant
bit_width = 3
narrow_range = True
zero_point_impl = ZeroZeroPoint
signed = True
proxy_class = ActQuantProxyFromInjector
And running this test:
quant = QuantIdentity(act_quant=ActInt)
x = torch.Tensor([-1, 2, 3])
y = quant(x)
print(y, y.is_valid)
Python version 3.6.9 and Brevitas 0.7.1
Is anything missing for my quantizer class? Usually I've encountered that the dependencies library throws an error if the given quantizer is not valid, but in this case an exception is thrown during the forward operation.