US 12,277,398 B2
Model training
Jian Gong, Beijing (CN); Yu Sun, Beijing (CN); Hao Tian, Beijing (CN); Hua Wu, Beijing (CN); Haifeng Wang, Beijing (CN); and Qiaoqiao She, Beijing (CN)
Assigned to Beijing Baidu Netcom Science and Technology Co., Ltd., Beijing (CN)
Filed by BEIJING BAIDU NETCOM SCIENCE TECHNOLOGY CO., LTD., Beijing (CN)
Filed on Mar. 14, 2022, as Appl. No. 17/694,034.
Claims priority of application No. 202110386608.2 (CN), filed on Apr. 12, 2021.
Prior Publication US 2022/0198153 A1, Jun. 23, 2022
Int. Cl. G06F 40/40 (2020.01); G06F 40/205 (2020.01); G06F 40/284 (2020.01)
CPC G06F 40/40 (2020.01) [G06F 40/205 (2020.01); G06F 40/284 (2020.01)] 20 Claims
OG exemplary drawing
 
1. A model training method for natural language processing, the method comprising:
receiving, by a computing device, an input through an interaction on an interactive interface performed using at least one interaction mode, wherein the at least one interaction mode is associated with the interactive interface;
determining, by the computing device, based on the input, a user-oriented prefabricated function, wherein the user-oriented prefabricated function is a text processing function associated with a model to be trained;
determining, by the computing device, based on the input, a model training function associated with a training process of the model, wherein the model training function includes one or more of the following functions: an actuator initializing function, an operational program initializing function, a single-machine-multi-card environment configuration function, a multi-machine-multi-card environment configuration function, a multi-machine central processing unit environment configuration function, a model loading function, or a model saving function;
determining, by the computing device, based on the input, a pre-trained model that is pre-trained based on deep learning;
determining, by the computing device, based on the input, a network structure associated with the pre-trained model;
providing, by the computing device, an initial network structure, wherein the initial network structure is a common structure integrated by a plurality of different reference pre-trained models;
configuring, by the computing device, the initial network structure as the network structure associated with the pre-trained model so as to support use of the pre-trained model;
training, by the computing device, based on the input, the model using the user-oriented prefabricated function, the model training function, and the pre-trained model;
providing, by the computing device, through the interaction performed using the at least one interaction mode, an output associated with the trained model; and
using, by the computing device, a function associated with applying the trained model so as to apply the trained model, wherein the function associated with applying the trained model includes providing one or more functions as follows: a special processing unit deployment function, a central processing unit deployment function, a single prediction function, a batch prediction function, a C++application programming interface function, or a Python application programming interface function.