Float8E4M3
Structs · AiDotNet.MixedPrecision
Represents an 8-bit floating point number in E4M3 format (4 exponent bits, 3 mantissa bits).
Inheritance: ValueType → Float8E4M3
Implements: IComparable<Float8E4M3>, IEquatable<Float8E4M3>
For Beginners
E4M3 is like a compressed version of regular floating point numbers. It uses only 8 bits (1 byte) instead of 32 bits, making it 4x smaller. The trade-off is reduced precision and range.
How It Works
Format Details:
- 1 sign bit
- 4 exponent bits (bias = 7)
- 3 mantissa bits
- Range: ±448
- Smallest positive: ~0.001953125
Use Cases: Best for weights and activations in neural network forward passes where values are typically well-bounded.
Hardware: NVIDIA H100/H200 GPUs have native FP8 Tensor Cores that can process E4M3 values at 2x the throughput of FP16.
Properties
| Property | Summary |
|---|---|
bool IsNaN { get; } |
Returns true if this value is NaN. |
bool IsNegative { get; } |
Returns true if this value is negative. |
bool IsZero { get; } |
Returns true if this value is zero. |
byte RawValue { get; } |
Gets the raw byte value. |
Methods
| Method | Summary |
|---|---|
Float8E4M3 FromFloat(float value) |
Creates a Float8E4M3 from a single-precision float. |
Float8E4M3 op_Explicit(float value) |
|
bool Equals(Float8E4M3 other) |
|
bool Equals(object obj) |
|
bool op_Equality(Float8E4M3 left, Float8E4M3 right) |
|
bool op_GreaterThan(Float8E4M3 left, Float8E4M3 right) |
|
bool op_GreaterThanOrEqual(Float8E4M3 left, Float8E4M3 right) |
|
bool op_Inequality(Float8E4M3 left, Float8E4M3 right) |
|
bool op_LessThan(Float8E4M3 left, Float8E4M3 right) |
|
bool op_LessThanOrEqual(Float8E4M3 left, Float8E4M3 right) |
|
float ToFloat() |
Converts this Float8E4M3 to a single-precision float. |
float op_Explicit(Float8E4M3 value) |
|
int CompareTo(Float8E4M3 other) |
|
int GetHashCode() |
|
string ToString() |
Fields
| Field | Summary |
|---|---|
Float8E4M3 MaxValue |
Maximum representable value in E4M3 format (448). |
Float8E4M3 MinPositive |
Minimum representable positive value in E4M3 format. |
Float8E4M3 NaN |
Represents NaN (Not a Number). |
Float8E4M3 One |
Represents positive one. |
Float8E4M3 Zero |
Represents positive zero. |