Summary of Rose: a Reward-oriented Data Selection Framework For Llm Task-specific Instruction Tuning, by Yang Wu et al.
ROSE: A Reward-Oriented Data Selection Framework for LLM Task-Specific Instruction Tuning
by Yang Wu, Huayi Zhang, Yizheng Jiao, Lin Ma, Xiaozhong Liu, Jinhong Yu, Dongyu Zhang, Dezhi Yu, Wei Xu
First submitted to arxiv on: 1 Dec 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The proposed ROSE method is a novel approach to selecting training data for task-specific instruction tuning of large language models (LLMs). The existing methods rely on similarity metrics, but this often leads to a misalignment between the instruction tuning loss and actual task performance. To address this issue, ROSE leverages pairwise preference loss as a reward signal to optimize data selection. Experimental results show that by selecting just 5% of the training data using ROSE, it can achieve competitive results compared to fine-tuning with the full dataset, and surpass state-of-the-art methods for task-specific instruction tuning. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models (LLMs) have shown great potential in producing more human-controllable outputs. This paper focuses on selecting the right training data for LLMs to improve performance on a specific task. Current methods use similarity metrics, but this can lead to poor results. The researchers created a new method called ROSE that uses a different approach to select the best data. By using just 5% of the available data, ROSE can perform as well as using all the data. |
Keywords
» Artificial intelligence » Fine tuning » Instruction tuning
