Summary of Communication-efficient and Tensorized Federated Fine-tuning Of Large Language Models, by Sajjad Ghiasvand et al.
Communication-Efficient and Tensorized Federated Fine-Tuning of Large Language Models
by Sajjad Ghiasvand, Yifan Yang, Zhiyu Xue, Mahnoosh Alizadeh, Zheng Zhang, Ramtin Pedarsani
First submitted to arxiv on: 16 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: 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 This paper introduces two novel fine-tuning methods, FedTT and FedTT+, for adapting Large Language Models (LLMs) in Federated Learning (FL) scenarios. These methods integrate tensorized adapters into client-side models’ encoder/decoder blocks to address the challenges of communication overhead and data heterogeneity. FedTT is applicable to both cross-silo FL and large-scale cross-device FL, while FedTT+ enhances robustness against data heterogeneity by adaptively freezing portions of tensor factors. Experiments on BERT and LLaMA models show that these methods successfully mitigate data heterogeneity challenges and perform competitively with existing federated PEFT approaches, achieving up to 10x reduction in communication cost. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research paper is about making language models work better when they’re trained on different devices. Normally, these models are trained on one device, but what if you want to use them for lots of different tasks? The problem is that each task has its own special data, and that makes it hard to train the model. The researchers created two new ways to make this work: FedTT and FedTT+. They tested their methods on some big language models and found that they worked really well. This means we can use these models for lots of different things without having to worry about how they were trained. |
Keywords
» Artificial intelligence » Bert » Encoder decoder » Federated learning » Fine tuning » Llama
