What is INT8 for Quantization?

luhengok · August 14, 2023, 9:08am

I am confused about how an int8 type could used for calaulating, especially all parameters are decimal.

gent.spah · August 14, 2023, 10:45am

You use only 8 bits repressing the number, you use them rounded up to integers or as far as I can remember you could use approximations of the floating number only with 8 bits, there is a technique I think (search in google). Check this link:

Topic		Replies	Views
Hi! Maybe a stupid project for a classification problem Introductions introductions , project	2	28	December 17, 2024
Rescale integer should be 255. or 255 in data augmentation assignment Convolutional Neural Networks in TensorFlow week-module-2	13	673	June 7, 2023
Are the weights in a neural network always of type Float? Neural Networks and Deep Learning ai-discussions , coursera-platform	6	512	January 17, 2024
Week 1: Computational challenges of training LLMs Generative AI with Large Language Models large-language-model , llm	3	40	November 18, 2024
Questions on quantizing both activation and weights for inference layers? Quantization In Depth	0	100	May 28, 2024

What is INT8 for Quantization?

Related topics