Neural Network Quantization Hardware Implementation

Neural Network Quantization Hardware Implementation

quantization operator

image-20230116205947777

Several different rounding methods

image-20230116210123095

image-20230116210234273

image-20230116210336096

quantization map

image-20230116210531386

image-20230116210655282

Quantization method

image-20230116211315877

image-20230116211431495

image-20230116212009174

image-20230116212014183

image-20230116213726209

The method corresponding to PPQ uses

image-20230116213900006

Quantitative calculation

image-20230116214032707

image-20230118210117161

image-20230118210254044

image-20230118211109781

image-20230118211123626

multiplication operator

image-20230118211156423

image-20230118211304443

image-20230118211511824

image-20230118211617703

image-20230118211903360

Addition operator

image-20230118211824592

image-20230118212043493

Activation function operator

image-20230118212210783

image-20230118212245145

image-20230118212304892

image-20230118212438411

image-20230118212510829

quantization matrix multiplication

image-20230118212706750

Operations sent to L2

image-20230118213139424

image-20230118213244005

image-20230118213303826

image-20230118213413735

nonlinear operator

image-20230118213522308

image-20230118213711599

image-20230118213809251

image-20230118213851909

image-20230118213938885

Summarize

image-20230118214039931

Guess you like

Origin blog.csdn.net/charles_zhang_/article/details/128730720