QuantizeTensorKwargs

class torchao.quantization.quantize_.common.QuantizeTensorKwargs

Base class for keyword argument container for quantized tensor creation. This is needed to support storing activation construction arguments on the weight tensor while supporting multiple types of activation quantization.

e.g.

class Float8Tensor(…)

@classmethod def to_float8(cls, tensor, quant_kwargs: QuantizeTensorKwargs)

…