Data processing method, device, computer equipment and storage medium

A neural network quantization method applied to a neural network quantization device which includes a control module and a processing module, wherein the processing module includes a first operation sub-module which includes a main operation sub-module and a secondary operation sub-module, the neural network quantization method comprising:

• determining, by the control module, a plurality pieces of data to be quantized from target data of a neural network, and obtaining a quantization result of the target data according to a piece of quantized data corresponding to each piece of data to be quantized, wherein the piece of quantized data corresponding to each piece of data to be quantized is obtained by quantizing the piece of data to be quantized by using a corresponding quantization parameter, wherein the quantization parameter includes a point location;
• and
• performing, by the first operation sub-module, an operation related to the quantization result to obtain an operation result, wherein performing, by the first operation sub-module, an operation related to the quantization result to obtain the operation result includes:
• sending, by the main operation sub-module, first data to the secondary operation sub-module, wherein the first data includes first-style data, which is in the quantization result and is obtained by quantizing according to the point location;
• performing, by the secondary operation sub-module, multiplication on the received first data to obtain an intermediate result;
• computing the corresponding quantization parameter according to each piece of data to be quantized and the corresponding data bit width;
• wherein computing the corresponding quantization parameter according to each piece of data to be quantized and the corresponding data bit width includes:
• when the quantization parameter does not include an offset, obtaining a first-kind point location of each piece of data to be quantized according to a maximum value of an absolute value of each piece of data to be quantized and the corresponding data bit width;
• when the quantization parameter does not include the offset, obtaining a maximum value of the piece of quantized data according to each piece of data to be quantized and the corresponding data bit width, and obtaining a first-kind scale factor of each piece of data to be quantized according to a maximum value of an absolute value of each piece of data to be quantized and the maximum value of the piece of quantized data;
• when the quantization parameter includes an offset, obtaining a second-kind point location of each piece of data to be quantized according to a maximum value and a minimum value of each piece of data to be quantized and the corresponding data bit width;
• when the quantization parameter includes an offset, obtaining a maximum value of the piece of quantized data according to each piece of data to be quantized and the corresponding data bit width, and obtaining a second-kind scale factor of each piece of data to be quantized according to a maximum value and a minimum value of each piece of data to be quantized as well as the maximum value of the piece of quantized data;
• or
• obtaining the offset of each piece of data to be quantized according to a maximum value and a minimum value in each piece of data to be quantized;
• performing, by the main operation sub-module, an operation on data except for the first data in the intermediate result and the quantization result to obtain the operation result;
• and
• data to be quantized is quantized according to the corresponding quantization parameter, to reduce a storage space of the data while ensuring a precision, and ensuring an accuracy and a reliability of the operation result and increase an operation efficiency.

The neural network quantization method of 9, further comprising:

• determining a quantization error corresponding to each piece of data to be quantized according to each piece of data to be quantized and the piece of quantized data corresponding to each piece of data to be quantized;
• adjusting the data bit width corresponding to each piece of data to be quantized to obtain an adjusted bit width corresponding to each piece of data to be quantized according to the quantization error and an error threshold corresponding to each piece of data to be quantized;
• and
• updating the data bit width corresponding to each piece of data to be quantized to the corresponding adjusted bit width, and computing a corresponding adjusted quantization parameter according to each piece of data to be quantized and the corresponding adjusted bit width to quantize each piece of data to be quantized according to the corresponding adjusted quantization parameter.