Summary of Accelerating Communication in Deep Learning Recommendation Model Training with Dual-level Adaptive Lossy Compression, by Hao Feng et al.
Accelerating Communication in Deep Learning Recommendation Model Training with Dual-Level Adaptive Lossy Compression
by Hao Feng, Boyuan Zhang, Fanjiang Ye, Min Si, Ching-Hsiang Chu, Jiannan Tian, Chunxing Yin, Summer Deng, Yuchen Hao, Pavan Balaji, Tong Geng, Dingwen Tao
First submitted to arxiv on: 5 Jul 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Distributed, Parallel, and Cluster Computing (cs.DC)
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 DLRM recommendation system model has gained popularity across various industries due to its state-of-the-art performance. However, training these large models on multiple devices or GPUs is time-consuming and requires significant communication resources. The authors introduce an error-bounded lossy compression method to reduce the communication data size and accelerate training. They develop a novel algorithm that achieves high compression ratios by analyzing embedding data features and adjusting error bounds using a dual-level adaptive strategy. The compressor is optimized for PyTorch tensors on GPUs, minimizing overhead. Evaluation shows a 1.38x training speedup with minimal accuracy impact. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary DLRM models are used to make recommendations in many industries, but they take a long time to train because they need to talk to each other. Researchers have developed a new way to compress the data that needs to be shared, making it faster and more efficient. They came up with an algorithm that can shrink the data without losing its accuracy. The new method is fast and doesn’t sacrifice performance. It’s like a superpower for computers! |
Keywords
* Artificial intelligence * Embedding
