Module tornado.api

Class QuantizationUtils

java.lang.Object
uk.ac.manchester.tornado.api.utils.QuantizationUtils

public class QuantizationUtils extends Object
Utility class providing operators for quantized data types. Currently, the functions implemented are compatible with 8-bit quantized models (e.g., Q8_0), but it will be extended as support is added for additional quantization formats and optimization strategies.
  • Constructor Details

    • QuantizationUtils

      public QuantizationUtils()
  • Method Details

    • dp4a

      public static int dp4a(Int8Array a, long offsetA, Int8Array b, long offsetB, int c)
    • dp4a

      public static int dp4a(Int8Array aGlobal, long offsetA, byte[] bLocal, long offsetB, int c)
    • dequantizeFusedResult

      public static float dequantizeFusedResult(int dotProductResult, float wScale, float xScale)
    • dp4a_packed

      public static int dp4a_packed(int a, int b, int c)