DoG-Instruct: Towards Premium Instruction-Tuning Data via Text-Grounded Instruction Wrapping
This addresses the problem of scalable, high-quality data generation for instruction-tuning in LLMs, offering a novel method to reduce costs and hallucinations.
The paper tackles the challenge of generating high-quality instruction-response pairs for LLMs by training LLMs to create these pairs from human-written documents, reducing hallucinations and bridging document styles. It results in a 10% relative improvement on AlpacaEval with only 1/5 of the training data compared to the best baseline.
The improvement of LLMs' instruction-following capabilities relies heavily on the availability of high-quality instruction-response pairs. Unfortunately, the current methods used to collect the pairs suffer from either unaffordable labor costs or severe hallucinations in the self-generation of LLM. To tackle these challenges, this paper proposes a scalable solution. It involves training LLMs to generate instruction-response pairs based on human-written documents, rather than relying solely on self-generation without context. Our proposed method not only exploits the advantages of human-written documents in reducing hallucinations but also utilizes an LLM to wrap the expression of documents, which enables us to bridge the gap between various document styles and the standard AI response. Experiments demonstrate that our method outperforms existing typical methods on multiple benchmarks. In particular, compared to the best-performing baseline, the LLM trained using our generated dataset exhibits a 10\% relative improvement in performance on AlpacaEval, despite utilizing only 1/5 of its training data. Furthermore, a comprehensive manual evaluation validates the quality of the data we generated. Our trained wrapper is publicly available at https://github.com/Bahuia/Dog-Instruct.