Papers › End-to-End Long Document Summarization using Gradient Caching

End-to-End Long Document Summarization using Gradient Caching

3 Jan 2025arXiv:2501.01805archive 2025-07-28

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 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.

PaperPDF

Code

No code repository is listed for this paper in the archive or in Syntology's graph.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

DecoderDocument SummarizationLong-Form Narrative Summarization

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Long-Form Narrative Summarization BookSum CachED (BART Large) BERTScore (F1) 54.4 #2 of 10 Archive leaderboard report
Long-Form Narrative Summarization BookSum SLED (BART Large) BERTScore (F1) 52.4 #3 of 10 Archive leaderboard report
Long-Form Narrative Summarization BookSum Unlimiformer (BART Base) BERTScore (F1) 51.5 #4 of 10 Archive leaderboard report
Long-Form Narrative Summarization BookSum Zero-Shot (GPT-4o) BERTScore (F1) 47.24 #5 of 10 Archive leaderboard report
Long-Form Narrative Summarization BookSum Zero-Shot (GPT-4o) ROUGE-1 20.3 #5 of 10 Archive leaderboard report
Long-Form Narrative Summarization BookSum Zero-Shot (GPT-4o) ROUGE-2 3.5 #5 of 10 Archive leaderboard report
Long-Form Narrative Summarization BookSum Zero-Shot (GPT-4o) ROUGE-L 17.68 #5 of 10 Archive leaderboard report
Long-Form Narrative Summarization MENSA CachED (BART Large) BERTScore (F1) 64.6 #2 of 10 Archive leaderboard report
Long-Form Narrative Summarization MENSA Unlimiformer (BART Base) BERTScore (F1) 58.7 #4 of 10 Archive leaderboard report
Long-Form Narrative Summarization MENSA SLED (BART Large) BERTScore (F1) 58.3 #5 of 10 Archive leaderboard report
Long-Form Narrative Summarization MENSA Zero-Shot (GPT-4o) BERTScore (F1) 52.8 #9 of 10 Archive leaderboard report
Long-Form Narrative Summarization SummScreen CachED (BART Large) BERTScore (F1) 61.59 #2 of 5 Archive leaderboard report
Long-Form Narrative Summarization SummScreen SLED (BART Large) BERTScore (F1) 59.9 #3 of 5 Archive leaderboard report
Long-Form Narrative Summarization SummScreen Unlimiformer (BART Base) BERTScore (F1) 58.5 #4 of 5 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

AdamAttentionBARTBPEDense ConnectionsDropoutLayer NormalizationLinear LayerMulti-Head AttentionResidual ConnectionSoftmax

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections