Summary of Collaboratively Adding New Knowledge to An Llm, by Rhui Dih Lee and Laura Wynter
Collaboratively adding new knowledge to an LLM
by Rhui Dih Lee, Laura Wynter
First submitted to arxiv on: 18 Oct 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 A novel approach is presented to successively add new knowledge to large language models (LLMs) while retaining previously acquired knowledge. The method, called LoRA, is compared to full-fine tuning of all parameters in various settings. In semi-cooperative scenarios where datasets are not available after training, LOE mixing, model merging, and LoRA-based orthogonal subspace sequential learning show promising results. In fully-cooperative settings where datasets remain available, joint training and sequential training with replay are effective approaches, with LoRA training generally outperforming full fine-tuning. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary A team of researchers developed a way to add new information to big language models while keeping the old knowledge they already had. They tested this method, called LoRA, against another approach that changes all the model’s parameters. In some cases where data isn’t available after training, certain combinations of techniques worked well. When data is available, joint training and sequential training with repetition are effective ways to add new information while keeping what was learned before. |
Keywords
» Artificial intelligence » Fine tuning » Lora
