SELECTLLM introduces a novel approach to selecting unlabelled instructions for annotation, leveraging LLM capabilities. The framework divides the dataset into subsets using clustering and prompts the LLM to identify beneficial instructions. Experimental results show SELECTLLM consistently outperforms other methods across different datasets and sample sizes. The framework demonstrates better cross-dataset generalization and qualitative response quality compared to baselines.
Іншою мовою
із вихідного контенту
arxiv.org
Ключові висновки, отримані з
by Ritik Sachin... о arxiv.org 03-07-2024
https://arxiv.org/pdf/2401.16553.pdfГлибші Запити