ArthModel: Enhance Arithmetic Skills to Large Language Model (2311.18609v1)

Abstract: With the great success of ChatGPT, the research of LLMs has become increasingly popular. However, the models have several limitations, such as toxicity and pool performance of arithmetic solving. Meanwhile, LLM may have some potential abilities that have yet to be exploited. In this paper, we choose a different way to enhance the arithmetic ability of LLM. We propose to train LLM to generate a postfix expression related to the arithmetic problem and incorporate it with small pretrained models. Moreover, this small model transfers the token embeddings into real dense numbers and invokes native functions of a deep learning platform to get the correct answer. To generate the final result, we propose prompt injection for adding the result outputs by the small model to LLM. This work provides different ways of thinking, training and using a LLM. The codes and models will be released at \url{https://github.com/eteced/arithmetic_finetuning_v1}.