Training method for chart generation large model, chart generation method and related device

Abstract

The invention provides a training method for a chart generation large model, a chart generation method and a related device, and relates to the field of artificial intelligence, in particular to the technical fields of large models, natural language processing, deep learning and the like. The training method for the graph generation large model comprises the steps that a training sample pair is obtained, the training sample pair comprises a subject sample, a reference SVG graph corresponding to the subject sample and at least two SVG graph samples, and the at least two SVG graph samples are SVG graphs of the same graph type and different layouts; performing chart generation processing on the input chart generation large model by using the training sample to obtain a prediction SVG chart output by the chart generation large model; and based on the prediction SVG chart and the reference SVG chart, fine tuning the charts to generate model parameters of the large model. According to the large chart generation model, the SVG charts which are diversified in layout, attractive and tidy in typesetting and reasonable and rich in content can be automatically generated according to themes.

Images