Method __init__

tools/convert/quant/quant.py:84–86

(self, weight)

            stochastic_rounding=0,
        )
        self.extra["convrot_groupsize"] = torch.tensor(convrot_groupsize, dtype=torch.int32, device=w.device)
        return w_q.contiguous(), params.scale.to(torch.float32).contiguous(), self.extra


@CONVERT_WEIGHT_REGISTER("fp8")
class QuantWeightFP8(QuantTemplate):