def get_linear_mod_weight(mod: nn.Module) -> torch.Tensor:
if isinstance(mod, nn.Linear):
return mod.weight.detach()
elif isinstance(mod, nni.LinearReLU):
return mod[0].weight.detach()
else:
return mod._weight_bias()[0] # type: ignore[operator]
def get_linear_mod_weight(mod: nn.Module) -> torch.Tensor:
# TODO(future PR): make more generic, handle everything
if isinstance(mod, nn.Linear):
return mod.weight.detach()
elif isinstance(mod, nni.LinearReLU):
return mod[0].weight.detach()
else:
return mod._weight_bias()[0] # type: ignore
def get_conv_mod_weight(mod: nn.Module) -> torch.Tensor:
if (isinstance(mod, nn.Conv1d) or isinstance(mod, nn.Conv2d)
or isinstance(mod, nn.Conv3d)):
return mod.weight.detach()
elif (isinstance(mod, nni.ConvReLU1d) or isinstance(mod, nni.ConvReLU2d)
or isinstance(mod, nni.ConvReLU3d)):
return mod[0].weight.detach()
else:
return mod._weight_bias()[0] # type: ignore[operator]
def get_conv_mod_weight(mod: nn.Module) -> torch.Tensor:
# TODO(future PR): handle QAT variants
if (isinstance(mod, nn.Conv1d) or isinstance(mod, nn.Conv2d)
or isinstance(mod, nn.Conv3d)):
return mod.weight.detach()
elif (isinstance(mod, nni.ConvReLU1d) or isinstance(mod, nni.ConvReLU2d)
or isinstance(mod, nni.ConvReLU3d)):
return mod[0].weight.detach()
else:
return mod._weight_bias()[0] # type: ignore
def mod_weight_bias_0(mod: nn.Module) -> torch.Tensor:
return mod._weight_bias()[0] # type: ignore[operator]
def get_conv_mod_weight(mod: nn.Module) -> torch.Tensor:
# TODO(future PR): make more generic, handle everything
if isinstance(mod, nn.Conv2d):
return mod.weight.detach()
else:
return mod._weight_bias()[0] # type: ignore
