QuantizedConv2d

class aimet_torch.nn.QuantizedConv2d(*args, **kwargs)

Quantized subclass of torch.nn.Conv2d

forward(self, input: torch.Tensor) → torch.Tensor

Quantized forward of torch.nn.Conv2d.

The input(s), parameter(s) (if any), and output(s) will be quantized with self.input_quantizers, self.param_quantizers, and self.output_quantizers respectively.

For more information, see QuantizationMixin.

compute_encodings()

Compute encodings with additional protection against bias overflow. HTP expects bias_scale = input_scale * weight_scale, which can cause bias overflow if bias_scale is too small. If bias overflow is expected, this function increases weight_scale until |bias / (input_scale * weight_scale)| <= 2**31