NEURAL NETWORK MODEL QUANTIZATION METHOD AND RELATED DEVICE THEREOF

A neural network model quantization method and a related device thereof, applied to the field of artificial intelligence. The method comprises: acquiring a neural network graph structure, the neural network graph structure comprising a plurality of operation nodes; inserting a plurality of quantizat...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: SUN, Fangxuan, ZHANG, Xiaowen, LIAN, Shuo, CHANG, Jing, ZHOU, Jun, LIANG, Xue, WANG, Chenxi
Format: Patent
Sprache:chi ; eng ; fre
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:A neural network model quantization method and a related device thereof, applied to the field of artificial intelligence. The method comprises: acquiring a neural network graph structure, the neural network graph structure comprising a plurality of operation nodes; inserting a plurality of quantization nodes in the neural network graph structure to obtain a quantized model graph structure; and training the quantized model graph structure according to sample data to obtain a quantized model, the size of the quantized model being smaller than the size of the neural network graph structure. The training comprises: quantizing input data of each quantization node by using each quantization node in the plurality of quantization nodes to obtain output data of each quantization node. L'invention porte sur un procédé de quantification de modèle de réseau neuronal et sur un dispositif associé, appliqués au domaine de l'intelligence artificielle. Le procédé consiste : à acquérir une structure de graphe de réseau neurona