Summary of Coba: Convergence Balancer For Multitask Finetuning Of Large Language Models, by Zi Gong et al.
CoBa: Convergence Balancer for Multitask Finetuning of Large Language Models
by Zi Gong, Hang Yu, Cong Liao, Bingchang Liu, Chaoyu Chen, Jianguo Li
First submitted to arxiv on: 9 Oct 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: 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 presents CoBa, a new multi-task learning (MTL) approach designed to manage task convergence balance with minimal computational overhead. The approach uses Relative Convergence Scores (RCS), Absolute Convergence Scores (ACS), and a Divergence Factor (DF) to dynamically adjust task weights during training, ensuring that all tasks converge at an even pace while mitigating individual task divergence. This results in improved performance of large language models (LLMs) by up to 13% relative to baselines. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary CoBa is a new way to make language models better at doing many tasks at once. It helps the model learn more quickly and correctly, which makes it useful for things like chatbots and virtual assistants. The approach uses special scores and numbers to make sure that all the different tasks are learning equally well. This makes the model perform better overall. |
Keywords
» Artificial intelligence » Multi task
