Microfloat
Microfloats.Microfloat — Type
Microfloat{S,E,M,V}A Microfloat type has S sign bits (between 0 and 1), E exponent bits (between 1 and 8), and M mantissa bits (between 0 and 7).
Finite
Microfloats.Finite — Type
IEEE 754-like
These types have IEEE 754-like Inf/NaN encodings, with Inf being represented as all 1s in the exponent and a significand of zero, and NaN being represented as all 1s in the exponent and a non-zero significand.
Microfloats.IEEE_754_like — Type
IEEE_754_likeMicrofloats.Float8_E3M4 — Type
Float8_E3M4Properties
- Bits:
1sign +3exponent +4significand (8total) - Variant:
IEEE_754_like - Has Inf:
true - Has NaN:
true - Max normal:
15.5 - Min normal:
0.25 - Max subnormal:
0.234375 - Min subnormal:
0.015625
Microfloats.Float8_E4M3 — Type
Float8_E4M3Properties
- Bits:
1sign +4exponent +3significand (8total) - Variant:
IEEE_754_like - Has Inf:
true - Has NaN:
true - Max normal:
240.0 - Min normal:
0.015625 - Max subnormal:
0.013671875 - Min subnormal:
0.001953125
Microfloats.Float8_E5M2 — Type
Float8_E5M2Properties
- Bits:
1sign +5exponent +2significand (8total) - Variant:
IEEE_754_like - Has Inf:
true - Has NaN:
true - Max normal:
57344.0 - Min normal:
6.103515625e-5 - Max subnormal:
4.57763671875e-5 - Min subnormal:
1.52587890625e-5
Microfloats.Float6_E2M3 — Type
Float6_E2M3Properties
- Bits:
1sign +2exponent +3significand (6total) - Variant:
IEEE_754_like - Has Inf:
true - Has NaN:
true - Max normal:
3.75 - Min normal:
1.0 - Max subnormal:
0.875 - Min subnormal:
0.125
Microfloats.Float6_E3M2 — Type
Float6_E3M2Properties
- Bits:
1sign +3exponent +2significand (6total) - Variant:
IEEE_754_like - Has Inf:
true - Has NaN:
true - Max normal:
14.0 - Min normal:
0.25 - Max subnormal:
0.1875 - Min subnormal:
0.0625
Microfloats.Float4_E2M1 — Type
Float4_E2M1Properties
- Bits:
1sign +2exponent +1significand (4total) - Variant:
IEEE_754_like - Has Inf:
true - Has NaN:
true - Max normal:
3.0 - Min normal:
1.0 - Max subnormal:
0.5 - Min subnormal:
0.5
Microscaling (MX)
Types from Open Compute Project Microscaling Formats (MX) Specification, with MX_E5M2 adhering to the IEEE 754-like encoding of Inf/NaN, whereas MXE4M3 and MXE8M0 have no Inf, and only one representation of NaN (excluding the sign bit), and the finite types MXE3M2, MXE2M3, and MX_E2M1 which have no Inf or NaNs.
Microfloats.MX — Type
MXMicrofloats.MX_E5M2 — Type
MX_E5M2Properties
- Bits:
1sign +5exponent +2significand (8total) - Variant:
MX - Has Inf:
true - Has NaN:
true - Max normal:
57344.0 - Min normal:
6.103515625e-5 - Max subnormal:
4.57763671875e-5 - Min subnormal:
1.52587890625e-5
Microfloats.MX_E4M3 — Type
MX_E4M3Properties
- Bits:
1sign +4exponent +3significand (8total) - Variant:
MX - Has Inf:
false - Has NaN:
true - Max normal:
448.0 - Min normal:
0.015625 - Max subnormal:
0.013671875 - Min subnormal:
0.001953125
Microfloats.MX_E3M2 — Type
MX_E3M2Properties
- Bits:
1sign +3exponent +2significand (6total) - Variant:
MX - Has Inf:
false - Has NaN:
false - Max normal:
28.0 - Min normal:
0.25 - Max subnormal:
0.1875 - Min subnormal:
0.0625
Microfloats.MX_E2M3 — Type
MX_E2M3Properties
- Bits:
1sign +2exponent +3significand (6total) - Variant:
MX - Has Inf:
false - Has NaN:
false - Max normal:
7.5 - Min normal:
1.0 - Max subnormal:
0.875 - Min subnormal:
0.125
Microfloats.MX_E2M1 — Type
MX_E2M1Properties
- Bits:
1sign +2exponent +1significand (4total) - Variant:
MX - Has Inf:
false - Has NaN:
false - Max normal:
6.0 - Min normal:
1.0 - Max subnormal:
0.5 - Min subnormal:
0.5
Microfloats.MX_E8M0 — Type
MX_E8M0Properties
- Bits:
0sign +8exponent +0significand (8total) - Variant:
MX - Has Inf:
false - Has NaN:
true - Max normal:
1.7014118346046923e38 - Min normal:
1.1754943508222875e-38 - Max subnormal:
N/A - Min subnormal:
N/A