End-to-End Long Document Summarization using Gradient Caching
Transactions of the Association for Computational Linguistics · 10.10.2025 · 13 · 1271-1297
DOI: 10.1162/tacl.a.40 ↗Kurz gefasst
Fachartikel in der Zeitschrift Transactions of the Association for Computational Linguistics, online erschienen am 10.10.2025. Verfasst von Rohit Saxena und 2 weiteren Autor:innen. Die Arbeit wurde bisher 3-mal zitiert (Crossref) und nennt 48 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Training transformer-based encoder-decoder models for long document summarization poses a significant challenge due to the quadratic memory consumption during training. Several approaches have been proposed to extend the input length at test time, but training with these approaches is still difficult, requiring truncation of input documents and causing a mismatch between training and test conditions. In this work, we propose CachED (Gradient Caching for Encoder-Decoder models), an approach that enables end-to-end training of existing transformer-based encoder-decoder models, using the entire document without truncation. Specifically, we apply non-overlapping sliding windows to input documents, followed by fusion in decoder. During backpropagation, the gradients are cached at the decoder and are passed through the encoder in chunks by re-computing the hidden vectors, similar to gradient checkpointing. In the experiments on long document summarization, we extend BART to CachED BART, processing more than 500K tokens during training and achieving superior performance without using any additional parameters.
Zitieren
APA
Saxena, R., Tang, H., & Keller, F. (2025). End-to-End Long Document Summarization using Gradient Caching. Transactions of the Association for Computational Linguistics, 13, 1271-1297. https://doi.org/10.1162/tacl.a.40
Vancouver
Saxena R, Tang H, Keller F. End-to-End Long Document Summarization using Gradient Caching. Transactions of the Association for Computational Linguistics. 2025;13:1271-1297. doi:10.1162/tacl.a.40
BibTeX
@article{Saxena2025,
author = {Saxena, Rohit and Tang, Hao and Keller, Frank},
title = {End-to-End Long Document Summarization using Gradient Caching},
journal = {Transactions of the Association for Computational Linguistics},
year = {2025},
volume = {13},
pages = {1271--1297},
doi = {10.1162/tacl.a.40}
}
EUVIMED