LinearBn1d

class torch.ao.nn.intrinsic.qat.modules.linear_fused.LinearBn1d(in_features, out_features, bias=True, eps=1e-05, momentum=0.1, freeze_bn=False, qconfig=None)

A LinearBn1d module is a module fused from Linear and BatchNorm1d, attached with FakeQuantize modules for weight, used in quantization aware training.

We combined the interface of torch.nn.Linear and :class:torch.nn.BatchNorm1d`.

Similar to torch.nn.Linear, with FakeQuantize modules initialized to default.

Variables:

• freeze_bn –

• weight_fake_quant – fake quant module for weight

classmethod from_float(mod, use_precomputed_fake_quant=False)

Create a qat module from a float module or qparams_dict

Parameters:

mod – A float module, either produced by torch.ao.quantization utilities or directly from the user.

train(mode=True)

Batchnorm’s training behavior is using the self.training flag. Prevent changing it if BN is frozen. This makes sure that calling model.train() on a model with a frozen BN will behave properly.