EmbeddingVector.QuantizedData

Kotlin |Java

@ExperimentalAppSearchApi
public final class EmbeddingVector.QuantizedData

Holds the pre-quantized embedding values along with the quantization parameters.

The original float values can be reconstructed from the quantized byte values using the formula: originalValue = (quantizedValue / scale) + minValue.

Summary

Public constructors
`QuantizedData( float minValue, float scale, @NonNull byte[] quantizedValues )` Creates a new `QuantizedData`.

Public methods
`boolean`	`equals(@Nullable Object o)`
`static @NonNull EmbeddingVector.QuantizedData`	`fromUnquantizedValues(@NonNull float[] values)` Creates a new `QuantizedData` by quantizing the provided float values to 8-bit integers.
`float`	`getMinValue()` Returns the minimum float value across all dimensions before quantization.
`@NonNull byte[]`	`getQuantizedValues()` Returns the pre-quantized int8 vector data.
`float`	`getScale()` Returns the scaling factor used to quantize the vector.
`int`	`hashCode()`

Public constructors

QuantizedData

Added in 1.2.0-alpha02

public QuantizedData(
    float minValue,
    float scale,
    @NonNull byte[] quantizedValues
)

Creates a new QuantizedData.

The original float values can be reconstructed from the quantized byte values using the formula: originalValue = (quantizedValue / scale) + minValue.

Parameters
`float minValue`	The minimum float value across all dimensions before quantization.
`float scale`	The scaling factor used to quantize the vector.
`@NonNull byte[] quantizedValues`	The pre-quantized int8 vector data.

Throws
`java.lang.IllegalArgumentException`	if `quantizedValues` is empty.

Public methods

equals

public boolean equals(@Nullable Object o)

fromUnquantizedValues

Added in 1.2.0-alpha02

public static @NonNull EmbeddingVector.QuantizedData fromUnquantizedValues(@NonNull float[] values)

Creates a new QuantizedData by quantizing the provided float values to 8-bit integers.

Storing 8-bit quantized values directly helps reduce memory and disk space usage by approximately 75% compared to storing full 32-bit floats.

Parameters
`@NonNull float[] values`	The float values to quantize.

Returns
`@NonNull EmbeddingVector.QuantizedData`	The `QuantizedData` created by quantizing the float values.

Throws
`java.lang.IllegalArgumentException`	if `values` is empty.

getMinValue

Added in 1.2.0-alpha02

public float getMinValue()

Returns the minimum float value across all dimensions before quantization.

getQuantizedValues

Added in 1.2.0-alpha02

public @NonNull byte[] getQuantizedValues()

Returns the pre-quantized int8 vector data.

getScale

Added in 1.2.0-alpha02

public float getScale()

Returns the scaling factor used to quantize the vector.

hashCode

public int hashCode()