def generate_target(self, _input: InputType, idx: int = 1, same: bool = False) -> torch.Tensor:
with torch.no_grad():
_output = self.get_logits(_input)
target = _output.argsort(dim=-1, descending=True)[:, idx]
if same:
target = repeat_to_batch(target.mode(dim=0)[0], len(_input))
return target
def generate_target(module: nn.Module,
_input: torch.Tensor,
idx: int = 1,
same: bool = False) -> torch.Tensor:
with torch.no_grad():
_output: torch.Tensor = module(_input)
target = _output.argsort(dim=-1, descending=True)[:, idx]
if same:
target = repeat_to_batch(target.mode(dim=0)[0], len(_input))
return target
def get_fake_seq(self, X: torch.Tensor) -> torch.Tensor:
if len(X.shape) == 4:
X = X[0]
noise = torch.normal(mean=0.0, std=1.0, size=X.shape, device=X.device)
fake_seq = X + self.dist * self.attack.sigma * noise
return repeat_to_batch(fake_seq, batch_size=self.fake_query_num)