torchao.dtypes

to_nf4
to_affine_quantized_intx
to_affine_quantized_intx_static
to_affine_quantized_floatx
to_affine_quantized_floatx_static
to_affine_quantized_fpx
NF4Tensor	用于将权重转换为 QLoRA NF4 格式的 NF4Tensor 类
AffineQuantizedTensor	仿射量化张量子类。仿射量化意味着我们使用仿射变换来量化浮点张量