def int_weight(self, x: torch.Tensor):
zero_hw_sentinel = getattr(self, ZERO_HW_SENTINEL_NAME)
quant_weight, scale, _ = self.tensor_quant(x, zero_hw_sentinel)
quant_weight = quant_weight / scale
quant_weight = round_ste(quant_weight)
quant_weight = quant_weight.int()
return quant_weight
def __int__(self):
return round_ste(self.tensor / self.scale)
def forward(self, x: torch.Tensor):
return round_ste(x)