Summary of Low-rank Finetuning For Llms: a Fairness Perspective, by Saswat Das et al.
Low-rank finetuning for LLMs: A fairness perspective
by Saswat Das, Marco Romanelli, Cuong Tran, Zarreen Reza, Bhavya Kailkhura, Ferdinando Fioretto
First submitted to arxiv on: 28 May 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 Low-rank approximation techniques have become a popular approach for fine-tuning Large Language Models (LLMs) due to their reduced computational and memory requirements. This paper investigates the effectiveness of these methods in capturing shifts in fine-tuning datasets from initial pre-trained data distributions. The results show that low-rank fine-tuning can fall short in learning such shifts, leading to non-negligible side effects. Specifically, this approach inadvertently preserves undesirable biases and toxic behaviors when used for toxicity mitigation or fair model development. Our comprehensive empirical evidence on various models, datasets, and tasks highlights the importance of careful evaluation to promote responsible LLMs development. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how a special way of training language models affects their behavior. The method is called low-rank fine-tuning, and it makes training faster and uses less memory. But this paper shows that this approach can actually make the model worse by keeping bad behaviors like toxicity or bias. This is important because these models are used for things like predicting what people will write next. We tested different language models, datasets, and tasks to see how well low-rank fine-tuning works. Our results show that we need to be careful when using this method to make sure the models don’t have bad side effects. |
Keywords
» Artificial intelligence » Fine tuning
