End-to-End Long Document Summarization using Gradient Caching

Rohit Saxena, Hao Tang, Frank Keller

Training transformer-based encoder-decoder models for long document summarization poses a significant challenge due to the quadratic memory consumption during training. Several app…