Big language model reasoning method and device, medium and computing equipment

The invention discloses a reasoning method and device of a large language model, a medium and computing equipment. The method comprises the steps that a reasoning instruction is taken; based on the key value cache of a preset character string template, generating the reasoning value of each element...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: YANG-GONG YIFAN, ZHANG YUEZHE, CHUANG XIAOMING, ZHENG HANXUN
Format: Patent
Sprache:chi ; eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The invention discloses a reasoning method and device of a large language model, a medium and computing equipment. The method comprises the steps that a reasoning instruction is taken; based on the key value cache of a preset character string template, generating the reasoning value of each element contained in the preset character string template; wherein the preset character string template comprises each element and a placeholder thereof, and each placeholder can indicate a generation rule of the corresponding element; and obtaining a current reasoning result based on the reasoning value of each element. According to the embodiment of the invention, the preset character string template containing the elements and the generation rules thereof is written according to the fixed format, and when secondary reasoning is carried out on the same reasoning target, the large language model can directly call the key value cache corresponding to the preset character string template generated in the first reasoning pro