Improving Language Model Reasoning with Self-motivated Learning

arXiv:2404.07017v1 Announce Type: new
Abstract: Large-scale high-quality training data is important for improving the performance of models. After trained with data that has rationales (reasoning steps), models gain reasoning capability. However, the dataset with high-quality rationales is relatively scarce due to the high annotation cost. To address this issue, we propose \textit{Self-motivated Learning} framework. The framework motivates the model itself to automatically generate rationales on existing datasets. Based on the inherent rank from correctness across multiple rationales, the model learns …

abstract annotation arxiv capability cost cs.ai cs.cl data dataset however improving issue language language model performance quality reasoning scale training training data type