Summary of Less: Selecting Influential Data For Targeted Instruction Tuning, by Mengzhou Xia et al.
LESS: Selecting Influential Data for Targeted Instruction Tuning
by Mengzhou Xia, Sadhika Malladi, Suchin Gururangan, Sanjeev Arora, Danqi Chen
First submitted to arxiv on: 6 Feb 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
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 paper proposes a new algorithm, LESS (Low-rank gradiEnt Similarity Search), which is designed to tackle targeted instruction tuning in large language models. This approach aims to develop specialized capabilities by selecting relevant data from extensive datasets. LESS adapts existing influence formulations to work with the Adam optimizer and variable-length instruction data. The algorithm constructs a reusable gradient datastore and selects examples based on their similarity to few-shot examples embodying specific capabilities. Experimental results show that training on a selected 5% of the data can outperform training on the full dataset across diverse downstream tasks. Additionally, the selected data is highly transferable, allowing smaller models to be used for selecting useful data for larger models and models from different families. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper talks about making special chatbots that can reason and understand things in a specific way. They want to find the right data to train these chatbots using an algorithm called LESS. This algorithm helps identify the most important data by looking at how similar it is to examples of the skill or capability they want the chatbot to have. The results show that this approach works really well and can even be used to help smaller models learn from bigger ones. |
Keywords
* Artificial intelligence * Few shot * Instruction tuning
