Microscaling
Documentation for Microscaling.
Microscaling.BlockFormat
Microscaling.MXFP4
Microscaling.MXFP6_E2M3
Microscaling.MXFP6_E3M2
Microscaling.MXFP8_E4M3
Microscaling.MXFP8_E5M2
Microscaling.MXINT8
Microscaling.NVFP4
Microscaling.BlockFormat
— TypeBlockFormat{E,S,k}
A block format specifies the element type, scale type, and number of elements per block.
Microscaling.MXFP4
— TypeMXFP4
MXFP4
is a microscaling format using FP4 elements (E2M1, no NaN/Inf), with E8M0 scale factors, each scaling contiguous element blocks of 32.
Microscaling.MXFP6_E2M3
— TypeMXFP6_E2M3
MXFP6_E2M3
is a microscaling format using FP6 elements (E2M3, no NaN/Inf), with E8M0 scale factors, each scaling contiguous element blocks of 32.
Microscaling.MXFP6_E3M2
— TypeMXFP6_E3M2
MXFP6_E3M2
is a microscaling format using FP6 elements (E3M2, no NaN/Inf), with E8M0 scale factors, each scaling contiguous element blocks of 32.
Microscaling.MXFP8_E4M3
— TypeMXFP8_E4M3
MXFP8_E4M3
is a microscaling format using FP8 elements (E4M3, including NaN), with E8M0 scale factors, each scaling contiguous element blocks of 32.
Microscaling.MXFP8_E5M2
— TypeMXFP8_E5M2
MXFP8_E5M2
is a microscaling format using FP8 elements (E5M2, including NaN/Inf), with E8M0 scale factors, each scaling contiguous element blocks of 32.
Microscaling.MXINT8
— TypeMXINT8
MXINT8
is a microscaling format using INT8 elements (Int8, scaled down by 64), with E8M0 scale factors, each scaling contiguous element blocks of 32.
Microscaling.NVFP4
— TypeNVFP4
NVFP4
is a microscaling format using FP4 elements (E2M1, no NaN/Inf), with E4M3 scale factors, each scaling contiguous element blocks of 16.