| 114 | |
| 115 | class IntegerQuantizer(BaseQuantizer): |
| 116 | def __init__(self, bit, symmetric, granularity, **kwargs): |
| 117 | super().__init__(bit, symmetric, granularity, **kwargs) |
| 118 | if "int_range" in self.kwargs: |
| 119 | self.qmin = self.kwargs["int_range"][0] |
| 120 | self.qmax = self.kwargs["int_range"][1] |
| 121 | else: |
| 122 | if self.sym: |
| 123 | self.qmin = -(2 ** (self.bit - 1)) |
| 124 | self.qmax = 2 ** (self.bit - 1) - 1 |
| 125 | else: |
| 126 | self.qmin = 0.0 |
| 127 | self.qmax = 2**self.bit - 1 |
| 128 | |
| 129 | self.qmin = torch.tensor(self.qmin) |
| 130 | self.qmax = torch.tensor(self.qmax) |
| 131 | self.dst_nbins = 2**bit |
| 132 | |
| 133 | def quant(self, tensor, scales, zeros, qmax, qmin): |
| 134 | tensor = torch.clamp(torch.round(tensor / scales) + zeros, qmin, qmax) |