Weight quantization method and device based on hybrid segment coding and server

By employing a hybrid segmented encoding weight quantization method, which segments the floating-point weights based on the Laplace distribution, the problem of large quantization errors in existing technologies is solved, achieving high-precision quantization and adaptive enhancement for neural networks.

CN120996117AActive Publication Date: 2025-11-21ZHEJIANG XINMAI SILICON CO LTD
4 Cites 2 Cited by

Patent Information

Application Number
CN202511517510.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-10-23
Publication Date
2025-11-21
Estimated Expiration
2045-10-23

AI Technical Summary

Technical Problem

Existing quantization methods tend to introduce significant quantization errors when there are large differences in the numerical distribution between channels, resulting in low quantization accuracy and an inability to effectively adapt to non-uniform weight distributions within channels.

Method used

A weight quantization method based on hybrid segmented coding is adopted. The floating-point weights are segmented and quantized by Laplace distribution, which divides them into the main peak region, the medium sparse region and the positive value interval. The multiplication and addition calculation is performed by using the flag bit and the multiplication coefficient to improve the quantization accuracy.

Benefits of technology

While maintaining the same compression ratio, it significantly reduces quantization error, improves the quantization accuracy of neural networks, and enhances their adaptability to non-uniform weight distribution within channels.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention provides a weight quantification method and device based on mixed segment coding and a server, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining a to-be-processed floating point weight and an activation value; based on the Laplacian distribution of the floating point weights, carrying out hybrid segmented coding processing on the floating point weights to obtain a coding weight set in a quantization weight coding format; and determining a multiplication coefficient and a shift value of the coding weight based on the identification bit, and carrying out multiplication and addition calculation processing on the activation value and the weight information in the data bit based on the multiplication coefficient and the shift value to obtain an output activation value corresponding to the coding weight. According to the method, the quantization precision of the neural network can be remarkably improved, and the quantization error is reduced while the compression ratio is not changed.
Need to check novelty before this filing date? Find Prior Art