numeric

package
v0.2.0 Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Mar 16, 2026 License: Apache-2.0 Imports: 5 Imported by: 0

Documentation

Overview

Package numeric provides precision types, arithmetic operations, and generic constraints for the Zerfoo ML framework. It serves as the foundation layer supporting float8, float16, float32, and float64 numeric types with IEEE 754 compliance.

Index

Constants

This section is empty.

Variables

This section is empty.

Functions

func Float8TestData 

func Float8TestData() struct {
	Add           []ArithmeticTestCase[float8.Float8]
	Mul           []ArithmeticTestCase[float8.Float8]
	Div           []ArithmeticTestCase[float8.Float8]
	Tanh          []UnaryTestCase[float8.Float8]
	Sigmoid       []UnaryTestCase[float8.Float8]
	LeakyReLU     []LeakyReLUTestCase[float8.Float8]
	LeakyReLUGrad []LeakyReLUTestCase[float8.Float8]
	Sum           []SumTestCase[float8.Float8]
}

Float8TestData provides common test data for float8 operations.

func Float16TestData 

func Float16TestData() struct {
	Add           []ArithmeticTestCase[float16.Float16]
	Mul           []ArithmeticTestCase[float16.Float16]
	Div           []ArithmeticTestCase[float16.Float16]
	Tanh          []UnaryTestCase[float16.Float16]
	Sigmoid       []UnaryTestCase[float16.Float16]
	LeakyReLU     []LeakyReLUTestCase[float16.Float16]
	LeakyReLUGrad []LeakyReLUTestCase[float16.Float16]
	Sum           []SumTestCase[float16.Float16]
}

Float16TestData provides common test data for float16 operations.

func Pack4BitSlice 

func Pack4BitSlice(values []uint8) ([]uint8, error)

Pack4BitSlice packs a slice of 4-bit values into uint8 array Input length must be even. Each pair of values is packed into one uint8.

func Pack4BitWeights 

func Pack4BitWeights(low4, high4 uint8) (uint8, error)

Pack4BitWeights packs two 4-bit values into a single uint8 This is used for MatMulNBits where weights are stored as 4-bit values

func TestArithmeticOp 

func TestArithmeticOp[T any](t *testing.T, opName string, op func(T, T) T, equal func(T, T) bool, tests []ArithmeticTestCase[T])

TestArithmeticOp tests a binary arithmetic operation.

func TestLeakyReLUOp 

func TestLeakyReLUOp[T any](t *testing.T, opName string, op func(T, float64) T, toFloat32 func(T) float32, tests []LeakyReLUTestCase[T])

TestLeakyReLUOp tests LeakyReLU operations with epsilon tolerance.

func TestSumOp 

func TestSumOp[T any](t *testing.T, op func([]T) T, toFloat32 func(T) float32, tests []SumTestCase[T])

TestSumOp tests sum operations with epsilon tolerance.

func TestUnaryOp 

func TestUnaryOp[T any](t *testing.T, opName string, op func(T) T, equal func(T, T) bool, tests []UnaryTestCase[T])

TestUnaryOp tests a unary operation.

func Unpack4BitSlice 

func Unpack4BitSlice(packed []uint8) []uint8

Unpack4BitSlice unpacks a slice of uint8 values into 4-bit weights Each uint8 contains two 4-bit values, so output length is 2x input length

func Unpack4BitWeights 

func Unpack4BitWeights(packed uint8) (uint8, uint8)

Unpack4BitWeights extracts two 4-bit values from a single uint8 Returns (low4, high4) where low4 is bits [0:3] and high4 is bits [4:7]

Types

type Arithmetic 

type Arithmetic[T any] interface {
	// Basic binary operations
	Add(a, b T) T
	Sub(a, b T) T
	Mul(a, b T) T
	Div(a, b T) T

	// Activation functions and their derivatives
	Tanh(x T) T
	Sigmoid(x T) T
	ReLU(x T) T
	LeakyReLU(x T, alpha float64) T
	TanhGrad(x T) T    // Derivative of Tanh
	SigmoidGrad(x T) T // Derivative of Sigmoid
	ReLUGrad(x T) T
	LeakyReLUGrad(x T, alpha float64) T

	// Conversion from standard types
	FromFloat32(f float32) T
	FromFloat64(f float64) T
	One() T

	// IsZero checks if a value is zero.
	IsZero(v T) bool

	// Abs returns the absolute value of x.
	Abs(x T) T
	// Sum returns the sum of all elements in the slice.
	Sum(s []T) T
	// Exp returns e**x.
	Exp(x T) T
	// Log returns the natural logarithm of x.
	Log(x T) T
	// Pow returns base**exponent.
	Pow(base, exponent T) T

	// Sqrt returns the square root of x.
	Sqrt(x T) T

	// GreaterThan returns true if a is greater than b.
	GreaterThan(a, b T) bool
}

Arithmetic defines a generic interface for all mathematical operations required by the compute engine. This allows the engine to be completely agnostic to the specific numeric type it is operating on.

type ArithmeticTestCase 

type ArithmeticTestCase[T any] struct {
	// contains filtered or unexported fields
}

ArithmeticTestCase represents a test case for arithmetic operations.

type BFloat16Ops 

type BFloat16Ops struct{}

BFloat16Ops provides the implementation of the Arithmetic interface for the float16.BFloat16 type.

func (BFloat16Ops) Abs 

func (ops BFloat16Ops) Abs(x float16.BFloat16) float16.BFloat16

func (BFloat16Ops) Add 

func (ops BFloat16Ops) Add(a, b float16.BFloat16) float16.BFloat16

func (BFloat16Ops) Div 

func (ops BFloat16Ops) Div(a, b float16.BFloat16) float16.BFloat16

func (BFloat16Ops) Exp 

func (ops BFloat16Ops) Exp(x float16.BFloat16) float16.BFloat16

func (BFloat16Ops) FromFloat32 

func (ops BFloat16Ops) FromFloat32(f float32) float16.BFloat16

func (BFloat16Ops) FromFloat64 

func (ops BFloat16Ops) FromFloat64(f float64) float16.BFloat16

func (BFloat16Ops) GreaterThan 

func (ops BFloat16Ops) GreaterThan(a, b float16.BFloat16) bool

func (BFloat16Ops) IsZero 

func (ops BFloat16Ops) IsZero(v float16.BFloat16) bool

func (BFloat16Ops) LeakyReLU 

func (ops BFloat16Ops) LeakyReLU(x float16.BFloat16, alpha float64) float16.BFloat16

func (BFloat16Ops) LeakyReLUGrad 

func (ops BFloat16Ops) LeakyReLUGrad(x float16.BFloat16, alpha float64) float16.BFloat16

func (BFloat16Ops) Log 

func (ops BFloat16Ops) Log(x float16.BFloat16) float16.BFloat16

func (BFloat16Ops) Mul 

func (ops BFloat16Ops) Mul(a, b float16.BFloat16) float16.BFloat16

func (BFloat16Ops) One 

func (ops BFloat16Ops) One() float16.BFloat16

func (BFloat16Ops) Pow 

func (ops BFloat16Ops) Pow(base, exponent float16.BFloat16) float16.BFloat16

func (BFloat16Ops) ReLU 

func (ops BFloat16Ops) ReLU(x float16.BFloat16) float16.BFloat16

func (BFloat16Ops) ReLUGrad 

func (ops BFloat16Ops) ReLUGrad(x float16.BFloat16) float16.BFloat16

func (BFloat16Ops) Sigmoid 

func (ops BFloat16Ops) Sigmoid(x float16.BFloat16) float16.BFloat16

func (BFloat16Ops) SigmoidGrad 

func (ops BFloat16Ops) SigmoidGrad(x float16.BFloat16) float16.BFloat16

func (BFloat16Ops) Sqrt 

func (ops BFloat16Ops) Sqrt(x float16.BFloat16) float16.BFloat16

func (BFloat16Ops) Sub 

func (ops BFloat16Ops) Sub(a, b float16.BFloat16) float16.BFloat16

func (BFloat16Ops) Sum 

func (ops BFloat16Ops) Sum(s []float16.BFloat16) float16.BFloat16

func (BFloat16Ops) Tanh 

func (ops BFloat16Ops) Tanh(x float16.BFloat16) float16.BFloat16

func (BFloat16Ops) TanhGrad 

func (ops BFloat16Ops) TanhGrad(x float16.BFloat16) float16.BFloat16

type Float8Ops 

type Float8Ops struct{}

Float8Ops provides the implementation of the Arithmetic interface for the float8.Float8 type.

func (Float8Ops) Abs 

func (ops Float8Ops) Abs(x float8.Float8) float8.Float8

Abs computes the absolute value of x.

func (Float8Ops) Add 

func (ops Float8Ops) Add(a, b float8.Float8) float8.Float8

Add performs element-wise addition.

func (Float8Ops) Div 

func (ops Float8Ops) Div(a, b float8.Float8) float8.Float8

Div performs element-wise division.

func (Float8Ops) Exp 

func (ops Float8Ops) Exp(x float8.Float8) float8.Float8

Exp computes the exponential of x.

func (Float8Ops) FromFloat32 

func (ops Float8Ops) FromFloat32(f float32) float8.Float8

FromFloat32 converts a float32 to a float8.Float8.

func (Float8Ops) FromFloat64 

func (ops Float8Ops) FromFloat64(f float64) float8.Float8

FromFloat64 converts a float64 to a float8.Float8.

func (Float8Ops) GreaterThan 

func (ops Float8Ops) GreaterThan(a, b float8.Float8) bool

GreaterThan checks if a is greater than b.

func (Float8Ops) IsZero 

func (ops Float8Ops) IsZero(v float8.Float8) bool

IsZero checks if the given float8.Float8 value is zero.

func (Float8Ops) LeakyReLU 

func (ops Float8Ops) LeakyReLU(x float8.Float8, alpha float64) float8.Float8

LeakyReLU computes the Leaky Rectified Linear Unit function.

func (Float8Ops) LeakyReLUGrad 

func (ops Float8Ops) LeakyReLUGrad(x float8.Float8, alpha float64) float8.Float8

LeakyReLUGrad computes the gradient of the Leaky Rectified Linear Unit function.

func (Float8Ops) Log 

func (ops Float8Ops) Log(x float8.Float8) float8.Float8

Log computes the natural logarithm of x.

func (Float8Ops) Mul 

func (ops Float8Ops) Mul(a, b float8.Float8) float8.Float8

Mul performs element-wise multiplication.

func (Float8Ops) One 

func (ops Float8Ops) One() float8.Float8

One returns a float8.Float8 with value 1.

func (Float8Ops) Pow 

func (ops Float8Ops) Pow(base, exponent float8.Float8) float8.Float8

Pow computes base raised to the power of exponent.

func (Float8Ops) ReLU 

func (ops Float8Ops) ReLU(x float8.Float8) float8.Float8

ReLU computes the Rectified Linear Unit function.

func (Float8Ops) ReLUGrad 

func (ops Float8Ops) ReLUGrad(x float8.Float8) float8.Float8

ReLUGrad computes the gradient of the Rectified Linear Unit function.

func (Float8Ops) Sigmoid 

func (ops Float8Ops) Sigmoid(x float8.Float8) float8.Float8

Sigmoid computes the sigmoid function of x.

func (Float8Ops) SigmoidGrad 

func (ops Float8Ops) SigmoidGrad(x float8.Float8) float8.Float8

SigmoidGrad computes the gradient of the sigmoid function.

func (Float8Ops) Sqrt 

func (ops Float8Ops) Sqrt(x float8.Float8) float8.Float8

Sqrt computes the square root of x.

func (Float8Ops) Sub 

func (ops Float8Ops) Sub(a, b float8.Float8) float8.Float8

Sub performs element-wise subtraction.

func (Float8Ops) Sum 

func (ops Float8Ops) Sum(s []float8.Float8) float8.Float8

Sum computes the sum of elements in a slice.

func (Float8Ops) Tanh 

func (ops Float8Ops) Tanh(x float8.Float8) float8.Float8

Tanh computes the hyperbolic tangent of x.

func (Float8Ops) TanhGrad 

func (ops Float8Ops) TanhGrad(x float8.Float8) float8.Float8

TanhGrad computes the gradient of the hyperbolic tangent function.

func (Float8Ops) ToFloat32 

func (ops Float8Ops) ToFloat32(t float8.Float8) float32

ToFloat32 converts a float8.Float8 to a float32.

type Float16Ops 

type Float16Ops struct{}

Float16Ops provides the implementation of the Arithmetic interface for the float16.Float16 type.

func (Float16Ops) Abs 

func (ops Float16Ops) Abs(x float16.Float16) float16.Float16

Abs computes the absolute value of x.

func (Float16Ops) Add 

func (ops Float16Ops) Add(a, b float16.Float16) float16.Float16

Add performs element-wise addition.

func (Float16Ops) Div 

func (ops Float16Ops) Div(a, b float16.Float16) float16.Float16

Div performs element-wise division.

func (Float16Ops) Exp 

func (ops Float16Ops) Exp(x float16.Float16) float16.Float16

Exp computes the exponential of x.

func (Float16Ops) FromFloat32 

func (ops Float16Ops) FromFloat32(f float32) float16.Float16

FromFloat32 converts a float32 to a float16.Float16.

func (Float16Ops) FromFloat64 

func (ops Float16Ops) FromFloat64(f float64) float16.Float16

FromFloat64 converts a float64 to a float16.Float16.

func (Float16Ops) GreaterThan 

func (ops Float16Ops) GreaterThan(a, b float16.Float16) bool

GreaterThan checks if a is greater than b.

func (Float16Ops) IsZero 

func (ops Float16Ops) IsZero(v float16.Float16) bool

IsZero checks if the given float16.Float16 value is zero.

func (Float16Ops) LeakyReLU 

func (ops Float16Ops) LeakyReLU(x float16.Float16, alpha float64) float16.Float16

LeakyReLU computes the Leaky Rectified Linear Unit function.

func (Float16Ops) LeakyReLUGrad 

func (ops Float16Ops) LeakyReLUGrad(x float16.Float16, alpha float64) float16.Float16

LeakyReLUGrad computes the gradient of the Leaky Rectified Linear Unit function.

func (Float16Ops) Log 

func (ops Float16Ops) Log(x float16.Float16) float16.Float16

Log computes the natural logarithm of x.

func (Float16Ops) Mul 

func (ops Float16Ops) Mul(a, b float16.Float16) float16.Float16

Mul performs element-wise multiplication.

func (Float16Ops) One 

func (ops Float16Ops) One() float16.Float16

One returns a float16.Float16 with value 1.

func (Float16Ops) Pow 

func (ops Float16Ops) Pow(base, exponent float16.Float16) float16.Float16

Pow computes base raised to the power of exponent.

func (Float16Ops) ReLU 

func (ops Float16Ops) ReLU(x float16.Float16) float16.Float16

ReLU computes the Rectified Linear Unit function.

func (Float16Ops) ReLUGrad 

func (ops Float16Ops) ReLUGrad(x float16.Float16) float16.Float16

ReLUGrad computes the gradient of the Rectified Linear Unit function.

func (Float16Ops) Sigmoid 

func (ops Float16Ops) Sigmoid(x float16.Float16) float16.Float16

Sigmoid computes the sigmoid function of x.

func (Float16Ops) SigmoidGrad 

func (ops Float16Ops) SigmoidGrad(x float16.Float16) float16.Float16

SigmoidGrad computes the gradient of the sigmoid function.

func (Float16Ops) Sqrt 

func (ops Float16Ops) Sqrt(x float16.Float16) float16.Float16

Sqrt computes the square root of x.

func (Float16Ops) Sub 

func (ops Float16Ops) Sub(a, b float16.Float16) float16.Float16

Sub performs element-wise subtraction.

func (Float16Ops) Sum 

func (ops Float16Ops) Sum(s []float16.Float16) float16.Float16

Sum computes the sum of elements in a slice.

func (Float16Ops) Tanh 

func (ops Float16Ops) Tanh(x float16.Float16) float16.Float16

Tanh computes the hyperbolic tangent of x.

func (Float16Ops) TanhGrad 

func (ops Float16Ops) TanhGrad(x float16.Float16) float16.Float16

TanhGrad computes the gradient of the hyperbolic tangent function.

func (Float16Ops) ToFloat32 

func (ops Float16Ops) ToFloat32(t float16.Float16) float32

ToFloat32 converts a float16.Float16 to a float32.

type Float32Ops 

type Float32Ops struct{}

Float32Ops provides the implementation of the Arithmetic interface for the float32 type.

func (Float32Ops) Abs 

func (ops Float32Ops) Abs(x float32) float32

Abs computes the absolute value of x.

func (Float32Ops) Add 

func (ops Float32Ops) Add(a, b float32) float32

Add performs element-wise addition.

func (Float32Ops) Div 

func (ops Float32Ops) Div(a, b float32) float32

Div performs element-wise division.

func (Float32Ops) Exp 

func (ops Float32Ops) Exp(x float32) float32

Exp computes the exponential of x.

func (Float32Ops) FromFloat32 

func (ops Float32Ops) FromFloat32(f float32) float32

FromFloat32 converts a float32 to a float32.

func (Float32Ops) FromFloat64 

func (ops Float32Ops) FromFloat64(f float64) float32

FromFloat64 converts a float64 to a float32.

func (Float32Ops) GreaterThan 

func (ops Float32Ops) GreaterThan(a, b float32) bool

GreaterThan checks if a is greater than b.

func (Float32Ops) IsZero 

func (ops Float32Ops) IsZero(v float32) bool

IsZero checks if the given float32 value is zero.

func (Float32Ops) LeakyReLU 

func (ops Float32Ops) LeakyReLU(x float32, alpha float64) float32

LeakyReLU computes the Leaky Rectified Linear Unit function.

func (Float32Ops) LeakyReLUGrad 

func (ops Float32Ops) LeakyReLUGrad(x float32, alpha float64) float32

LeakyReLUGrad computes the gradient of the Leaky Rectified Linear Unit function.

func (Float32Ops) Log 

func (ops Float32Ops) Log(x float32) float32

Log computes the natural logarithm of x.

func (Float32Ops) Mul 

func (ops Float32Ops) Mul(a, b float32) float32

Mul performs element-wise multiplication.

func (Float32Ops) One 

func (ops Float32Ops) One() float32

One returns a float32 with value 1.

func (Float32Ops) Pow 

func (ops Float32Ops) Pow(base, exponent float32) float32

Pow computes base raised to the power of exponent.

func (Float32Ops) ReLU 

func (ops Float32Ops) ReLU(x float32) float32

ReLU computes the Rectified Linear Unit function.

func (Float32Ops) ReLUGrad 

func (ops Float32Ops) ReLUGrad(x float32) float32

ReLUGrad computes the gradient of the Rectified Linear Unit function.

func (Float32Ops) Sigmoid 

func (ops Float32Ops) Sigmoid(x float32) float32

Sigmoid computes the sigmoid function of x.

func (Float32Ops) SigmoidGrad 

func (ops Float32Ops) SigmoidGrad(x float32) float32

SigmoidGrad computes the gradient of the sigmoid function.

func (Float32Ops) Sqrt 

func (ops Float32Ops) Sqrt(x float32) float32

Sqrt computes the square root of x.

func (Float32Ops) Sub 

func (ops Float32Ops) Sub(a, b float32) float32

Sub performs element-wise subtraction.

func (Float32Ops) Sum 

func (ops Float32Ops) Sum(s []float32) float32

Sum computes the sum of elements in a slice.

func (Float32Ops) Tanh 

func (ops Float32Ops) Tanh(x float32) float32

Tanh computes the hyperbolic tangent of x.

func (Float32Ops) TanhGrad 

func (ops Float32Ops) TanhGrad(x float32) float32

TanhGrad computes the gradient of the hyperbolic tangent function.

func (Float32Ops) ToFloat32 

func (ops Float32Ops) ToFloat32(t float32) float32

ToFloat32 converts a float32 to a float32.

type Float64Ops 

type Float64Ops struct{}

Float64Ops provides the implementation of the Arithmetic interface for the float64 type.

func (Float64Ops) Abs 

func (ops Float64Ops) Abs(x float64) float64

Abs computes the absolute value of x.

func (Float64Ops) Add 

func (ops Float64Ops) Add(a, b float64) float64

Add performs element-wise addition.

func (Float64Ops) Div 

func (ops Float64Ops) Div(a, b float64) float64

Div performs element-wise division.

func (Float64Ops) Exp 

func (ops Float64Ops) Exp(x float64) float64

Exp computes the exponential of x.

func (Float64Ops) FromFloat32 

func (ops Float64Ops) FromFloat32(f float32) float64

FromFloat32 converts a float32 to a float64.

func (Float64Ops) FromFloat64 

func (ops Float64Ops) FromFloat64(f float64) float64

FromFloat64 converts a float64 to a float64.

func (Float64Ops) GreaterThan 

func (ops Float64Ops) GreaterThan(a, b float64) bool

GreaterThan checks if a is greater than b.

func (Float64Ops) IsZero 

func (ops Float64Ops) IsZero(v float64) bool

IsZero checks if the given float64 value is zero.

func (Float64Ops) LeakyReLU 

func (ops Float64Ops) LeakyReLU(x, alpha float64) float64

LeakyReLU computes the Leaky Rectified Linear Unit function.

func (Float64Ops) LeakyReLUGrad 

func (ops Float64Ops) LeakyReLUGrad(x, alpha float64) float64

LeakyReLUGrad computes the gradient of the Leaky Rectified Linear Unit function.

func (Float64Ops) Log 

func (ops Float64Ops) Log(x float64) float64

Log computes the natural logarithm of x.

func (Float64Ops) Mul 

func (ops Float64Ops) Mul(a, b float64) float64

Mul performs element-wise multiplication.

func (Float64Ops) One 

func (ops Float64Ops) One() float64

One returns a float64 with value 1.

func (Float64Ops) Pow 

func (ops Float64Ops) Pow(base, exponent float64) float64

Pow computes base raised to the power of exponent.

func (Float64Ops) ReLU 

func (ops Float64Ops) ReLU(x float64) float64

ReLU computes the Rectified Linear Unit function.

func (Float64Ops) ReLUGrad 

func (ops Float64Ops) ReLUGrad(x float64) float64

ReLUGrad computes the gradient of the Rectified Linear Unit function.

func (Float64Ops) Sigmoid 

func (ops Float64Ops) Sigmoid(x float64) float64

Sigmoid computes the sigmoid function of x.

func (Float64Ops) SigmoidGrad 

func (ops Float64Ops) SigmoidGrad(x float64) float64

SigmoidGrad computes the gradient of the sigmoid function.

func (Float64Ops) Sqrt 

func (ops Float64Ops) Sqrt(x float64) float64

Sqrt computes the square root of x.

func (Float64Ops) Sub 

func (ops Float64Ops) Sub(a, b float64) float64

Sub performs element-wise subtraction.

func (Float64Ops) Sum 

func (ops Float64Ops) Sum(s []float64) float64

Sum computes the sum of elements in a slice.

func (Float64Ops) Tanh 

func (ops Float64Ops) Tanh(x float64) float64

Tanh computes the hyperbolic tangent of x.

func (Float64Ops) TanhGrad 

func (ops Float64Ops) TanhGrad(x float64) float64

TanhGrad computes the gradient of the hyperbolic tangent function.

func (Float64Ops) ToFloat32 

func (ops Float64Ops) ToFloat32(t float64) float32

ToFloat32 converts a float64 to a float32.

type Int8Ops 

type Int8Ops struct{}

Int8Ops provides the implementation of the Arithmetic interface for the int8 type.

func (Int8Ops) Abs 

func (ops Int8Ops) Abs(x int8) int8

Abs computes the absolute value of x.

func (Int8Ops) Add 

func (ops Int8Ops) Add(a, b int8) int8

Add performs element-wise addition.

func (Int8Ops) Div 

func (ops Int8Ops) Div(a, b int8) int8

Div performs element-wise division.

func (Int8Ops) Exp 

func (ops Int8Ops) Exp(x int8) int8

Exp computes the exponential of x.

func (Int8Ops) FromFloat32 

func (ops Int8Ops) FromFloat32(f float32) int8

FromFloat32 converts a float32 to an int8.

func (Int8Ops) FromFloat64 

func (ops Int8Ops) FromFloat64(f float64) int8

FromFloat64 converts a float64 to an int8.

func (Int8Ops) GreaterThan 

func (ops Int8Ops) GreaterThan(a, b int8) bool

GreaterThan checks if a is greater than b.

func (Int8Ops) IsZero 

func (ops Int8Ops) IsZero(v int8) bool

IsZero checks if the given int8 value is zero.

func (Int8Ops) LeakyReLU 

func (ops Int8Ops) LeakyReLU(x int8, alpha float64) int8

LeakyReLU computes the Leaky Rectified Linear Unit function.

func (Int8Ops) LeakyReLUGrad 

func (ops Int8Ops) LeakyReLUGrad(x int8, alpha float64) int8

LeakyReLUGrad computes the gradient of the Leaky Rectified Linear Unit function.

func (Int8Ops) Log 

func (ops Int8Ops) Log(x int8) int8

Log computes the natural logarithm of x.

func (Int8Ops) Mul 

func (ops Int8Ops) Mul(a, b int8) int8

Mul performs element-wise multiplication.

func (Int8Ops) One 

func (ops Int8Ops) One() int8

One returns an int8 with value 1.

func (Int8Ops) Pow 

func (ops Int8Ops) Pow(base, exponent int8) int8

Pow computes base raised to the power of exponent.

func (Int8Ops) ReLU 

func (ops Int8Ops) ReLU(x int8) int8

ReLU computes the Rectified Linear Unit function.

func (Int8Ops) ReLUGrad 

func (ops Int8Ops) ReLUGrad(x int8) int8

ReLUGrad computes the gradient of the Rectified Linear Unit function.

func (Int8Ops) Sigmoid 

func (ops Int8Ops) Sigmoid(x int8) int8

Sigmoid computes the sigmoid function of x.

func (Int8Ops) SigmoidGrad 

func (ops Int8Ops) SigmoidGrad(x int8) int8

SigmoidGrad computes the gradient of the sigmoid function.

func (Int8Ops) Sqrt 

func (ops Int8Ops) Sqrt(x int8) int8

Sqrt computes the square root of x.

func (Int8Ops) Sub 

func (ops Int8Ops) Sub(a, b int8) int8

Sub performs element-wise subtraction.

func (Int8Ops) Sum 

func (ops Int8Ops) Sum(s []int8) int8

Sum computes the sum of elements in a slice.

func (Int8Ops) Tanh 

func (ops Int8Ops) Tanh(x int8) int8

Tanh computes the hyperbolic tangent of x.

func (Int8Ops) TanhGrad 

func (ops Int8Ops) TanhGrad(x int8) int8

TanhGrad computes the gradient of the hyperbolic tangent function.

type IntOps 

type IntOps struct{}

IntOps implements Arithmetic for int.

func (IntOps) Abs 

func (IntOps) Abs(x int) int

Abs computes the absolute value of x.

func (IntOps) Add 

func (IntOps) Add(a, b int) int

Add performs element-wise addition.

func (IntOps) Div 

func (IntOps) Div(a, b int) int

Div performs element-wise division.

func (IntOps) Exp 

func (IntOps) Exp(x int) int

Exp computes the exponential of x.

func (IntOps) FromFloat32 

func (IntOps) FromFloat32(f float32) int

FromFloat32 converts a float32 to an int.

func (IntOps) FromFloat64 

func (IntOps) FromFloat64(f float64) int

FromFloat64 converts a float64 to an int.

func (IntOps) GreaterThan 

func (IntOps) GreaterThan(a, b int) bool

GreaterThan checks if a is greater than b.

func (IntOps) IsZero 

func (IntOps) IsZero(v int) bool

IsZero checks if the given int value is zero.

func (IntOps) LeakyReLU 

func (IntOps) LeakyReLU(x int, alpha float64) int

LeakyReLU computes the Leaky Rectified Linear Unit function.

func (IntOps) LeakyReLUGrad 

func (IntOps) LeakyReLUGrad(x int, alpha float64) int

LeakyReLUGrad computes the gradient of the Leaky Rectified Linear Unit function.

func (IntOps) Log 

func (IntOps) Log(x int) int

Log computes the natural logarithm of x.

func (IntOps) Mul 

func (IntOps) Mul(a, b int) int

Mul performs element-wise multiplication.

func (IntOps) One 

func (IntOps) One() int

One returns an int with value 1.

func (IntOps) Pow 

func (IntOps) Pow(base, exponent int) int

Pow computes base raised to the power of exponent.

func (IntOps) ReLU 

func (IntOps) ReLU(x int) int

ReLU computes the Rectified Linear Unit function.

func (IntOps) ReLUGrad 

func (IntOps) ReLUGrad(x int) int

ReLUGrad computes the gradient of the Rectified Linear Unit function.

func (IntOps) Sigmoid 

func (IntOps) Sigmoid(x int) int

Sigmoid computes the sigmoid function of x.

func (IntOps) SigmoidGrad 

func (IntOps) SigmoidGrad(x int) int

SigmoidGrad computes the gradient of the sigmoid function.

func (IntOps) Sqrt 

func (IntOps) Sqrt(x int) int

Sqrt computes the square root of x.

func (IntOps) Sub 

func (IntOps) Sub(a, b int) int

Sub performs element-wise subtraction.

func (IntOps) Sum 

func (IntOps) Sum(s []int) int

Sum computes the sum of elements in a slice.

func (IntOps) Tanh 

func (IntOps) Tanh(x int) int

Tanh computes the hyperbolic tangent of x.

func (IntOps) TanhGrad 

func (IntOps) TanhGrad(x int) int

TanhGrad computes the gradient of the hyperbolic tangent function.

func (IntOps) ToFloat32 

func (IntOps) ToFloat32(t int) float32

ToFloat32 converts an int to a float32.

type LeakyReLUTestCase 

type LeakyReLUTestCase[T any] struct {
	// contains filtered or unexported fields
}

LeakyReLUTestCase represents a test case for LeakyReLU operations.

type QuantizationConfig 

type QuantizationConfig struct {
	Scale     float32
	ZeroPoint int64
	Symmetric bool // If true, zero_point is ignored and assumed to be 0
}

QuantizationConfig holds parameters for quantization operations

func ComputeQuantizationParams 

func ComputeQuantizationParams(minVal, maxVal float32, symmetric bool) (*QuantizationConfig, error)

ComputeQuantizationParams computes optimal scale and zero_point for a given data range This is useful for dynamic quantization where parameters aren't predetermined

func NewQuantizationConfig 

func NewQuantizationConfig(scale float32, zeroPoint int64, symmetric bool) (*QuantizationConfig, error)

NewQuantizationConfig creates a quantization configuration with validation

func (*QuantizationConfig) Dequantize 

func (qc *QuantizationConfig) Dequantize(quantized uint8) float32

Dequantize converts uint8 values back to float32 using linear dequantization Formula: dequantized = scale * (quantized - zero_point)

func (*QuantizationConfig) Dequantize4BitWeights 

func (qc *QuantizationConfig) Dequantize4BitWeights(packed []uint8) []float32

Dequantize4BitWeights combines 4-bit unpacking with dequantization This is the typical operation for MatMulNBits: unpack 4-bit -> dequantize -> float32

func (*QuantizationConfig) DequantizeSlice 

func (qc *QuantizationConfig) DequantizeSlice(quantized []uint8) []float32

DequantizeSlice dequantizes a slice of uint8 values to float32

func (*QuantizationConfig) QuantizationError 

func (qc *QuantizationConfig) QuantizationError(original []float32) float64

QuantizationError computes the quantization error (RMS) between original and quantized values

func (*QuantizationConfig) Quantize 

func (qc *QuantizationConfig) Quantize(value float32) uint8

Quantize converts float32 values to uint8 using linear quantization Formula: quantized = round(value / scale + zero_point)

func (*QuantizationConfig) QuantizeSlice 

func (qc *QuantizationConfig) QuantizeSlice(values []float32) []uint8

QuantizeSlice quantizes a slice of float32 values to uint8

func (*QuantizationConfig) ValidateQuantizationRoundTrip 

func (qc *QuantizationConfig) ValidateQuantizationRoundTrip(values []float32, tolerance float64) error

ValidateQuantizationRoundTrip checks that quantize->dequantize preserves values within tolerance

type SumTestCase 

type SumTestCase[T any] struct {
	// contains filtered or unexported fields
}

SumTestCase represents a test case for sum operations.

type Uint8Ops 

type Uint8Ops struct{}

Uint8Ops provides the implementation of the Arithmetic interface for the uint8 type.

func (Uint8Ops) Abs 

func (ops Uint8Ops) Abs(x uint8) uint8

Abs computes the absolute value of x.

func (Uint8Ops) Add 

func (ops Uint8Ops) Add(a, b uint8) uint8

Add performs element-wise addition.

func (Uint8Ops) Div 

func (ops Uint8Ops) Div(a, b uint8) uint8

Div performs element-wise division.

func (Uint8Ops) Exp 

func (ops Uint8Ops) Exp(x uint8) uint8

Exp computes the exponential of x.

func (Uint8Ops) FromFloat32 

func (ops Uint8Ops) FromFloat32(f float32) uint8

FromFloat32 converts a float32 to a uint8.

func (Uint8Ops) FromFloat64 

func (ops Uint8Ops) FromFloat64(f float64) uint8

FromFloat64 converts a float64 to a uint8.

func (Uint8Ops) GreaterThan 

func (ops Uint8Ops) GreaterThan(a, b uint8) bool

GreaterThan checks if a is greater than b.

func (Uint8Ops) IsZero 

func (ops Uint8Ops) IsZero(v uint8) bool

IsZero checks if the given uint8 value is zero.

func (Uint8Ops) LeakyReLU 

func (ops Uint8Ops) LeakyReLU(x uint8, alpha float64) uint8

LeakyReLU computes the Leaky Rectified Linear Unit function.

func (Uint8Ops) LeakyReLUGrad 

func (ops Uint8Ops) LeakyReLUGrad(x uint8, alpha float64) uint8

LeakyReLUGrad computes the gradient of the Leaky Rectified Linear Unit function.

func (Uint8Ops) Log 

func (ops Uint8Ops) Log(x uint8) uint8

Log computes the natural logarithm of x.

func (Uint8Ops) Mul 

func (ops Uint8Ops) Mul(a, b uint8) uint8

Mul performs element-wise multiplication.

func (Uint8Ops) One 

func (ops Uint8Ops) One() uint8

One returns a uint8 with value 1.

func (Uint8Ops) Pow 

func (ops Uint8Ops) Pow(base, exponent uint8) uint8

Pow computes base raised to the power of exponent.

func (Uint8Ops) ReLU 

func (ops Uint8Ops) ReLU(x uint8) uint8

ReLU computes the Rectified Linear Unit function.

func (Uint8Ops) ReLUGrad 

func (ops Uint8Ops) ReLUGrad(x uint8) uint8

ReLUGrad computes the gradient of the Rectified Linear Unit function.

func (Uint8Ops) Sigmoid 

func (ops Uint8Ops) Sigmoid(x uint8) uint8

Sigmoid computes the sigmoid function of x.

func (Uint8Ops) SigmoidGrad 

func (ops Uint8Ops) SigmoidGrad(x uint8) uint8

SigmoidGrad computes the gradient of the sigmoid function.

func (Uint8Ops) Sqrt 

func (ops Uint8Ops) Sqrt(x uint8) uint8

Sqrt computes the square root of x.

func (Uint8Ops) Sub 

func (ops Uint8Ops) Sub(a, b uint8) uint8

Sub performs element-wise subtraction.

func (Uint8Ops) Sum 

func (ops Uint8Ops) Sum(s []uint8) uint8

Sum computes the sum of elements in a slice.

func (Uint8Ops) Tanh 

func (ops Uint8Ops) Tanh(x uint8) uint8

Tanh computes the hyperbolic tangent of x.

func (Uint8Ops) TanhGrad 

func (ops Uint8Ops) TanhGrad(x uint8) uint8

TanhGrad computes the gradient of the hyperbolic tangent function.

type UnaryTestCase 

type UnaryTestCase[T any] struct {
	// contains filtered or unexported fields
}

UnaryTestCase represents a test case for unary operations.

Source Files

View all Source files

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.