bitorch.quantizations.identity.Identity

class bitorch.quantizations.identity.Identity

Module that provides the identity function, which can be useful for certain training strategies

Methods

quantize

forwards the input tensor x without quantization.

Attributes

bit_width
name

quantize(x: Tensor) → Tensor

forwards the input tensor x without quantization.

Parameters:

x (torch.Tensor) – tensor to be forwarded.

Returns:

tensor x

Return type:

torch.Tensor