首页|生成式预训练Transformer模型的逻辑性优化方法

生成式预训练Transformer模型的逻辑性优化方法

扫码查看
生成式预训练 Transformer(Generative Pre-Trained Transformer,GPT)模型作为一种基于Transformer架构的预训练模型,在完成自然语言处理任务方面取得了巨大的成功.由于依赖于生成下一个词的局部贪婪过程,使对任务或输出的全局理解、逻辑推理和道德法规约束能力不够.为了提升计算的逻辑性和可靠性,结合的生成型计算过程,论述计算结果的逻辑局限性,从而引入一类和逻辑计算模型混合的优化结构.
A Logic Optimization Method for Generative Pre-Trained Transformer Model
As a Pre-Trained model based on Transformer architecture,the Generative pre-trained Transformer model has achieved great success in completing natural language processing tasks.Due to the dependence of the Generative Pre-Trained Transformer(GPT)model on the local greedy process of generating the next word,the GPT model lacks global understanding,logical reasoning,and ethical constraints on the task or output.In order to improve the logic and reliability of GPT model calculations,the logical limitations of GPT model calculation results are discussed combined with the process of GPT model generation calculation,and a kind of optimization structure of GPT model and logical calculation model is introduced.

Generative Pre-Trained Transformer(GPT)logicalityoptimize structure

张兆天

展开 >

中央广播电视总台技术局,北京 100000

生成式预训练Transformer模型(GPT) 逻辑性 优化结构

2024

信息与电脑
北京电子控股有限责任公司

信息与电脑

影响因子:1.143
ISSN:1003-9767
年,卷(期):2024.36(4)
  • 5