Georgia Tech and Facebook Reduce Size of Deep Learning Recommendation Models by Factor of 112
Georgia Tech and Facebook AI Researchers Devise a New Tensor Train Approach to Reduce the Size of Deep Learning Recommendation Models up to 112x
TT-Rec: Tensor Train Compression for Deep Learning Recommendation Model Embeddings
Georgia Tech and Facebook arXiv paper https://arxiv.org/pdf/2101.11714.pdf
Facebook AI Research FBTT-Embedding GitHub
FBTT-Embedding library provides functionality to compress sparse embedding tables commonly usedin machine learning models such as recommendation and natural language processing.