EntropyLong: Effective Long-Context Training via Predictive Uncertainty

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Jia, Junlong, Chen, Ziyang, Wu, Xing, Gao, Chaochen, Lin, Zijia, Zhang, Debing, Hu, Songlin, Guo, Binghui
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866916985775849472
author Jia, Junlong
Chen, Ziyang
Wu, Xing
Gao, Chaochen
Lin, Zijia
Zhang, Debing
Hu, Songlin
Guo, Binghui
author_facet Jia, Junlong
Chen, Ziyang
Wu, Xing
Gao, Chaochen
Lin, Zijia
Zhang, Debing
Hu, Songlin
Guo, Binghui
contents Training long-context language models to capture long-range dependencies requires specialized data construction. Current approaches, such as generic text concatenation or heuristic-based variants, frequently fail to guarantee genuine long-range dependencies. We propose EntropyLong, a novel data construction method that leverages predictive uncertainty to verify dependency quality. Our approach identifies high-entropy positions in documents, retrieves semantically relevant contexts from large corpora, and verifies their utility by assessing whether they reduce prediction entropy. This model-in-the-loop verification ensures each dependency represents measurable information gain rather than spurious correlation. We construct training samples with long-range dependencies by combining original documents with these verified contextual supplements. Using FineWebEdu and Cosmopedia, we generate a dataset of 128K-length sequences with verified dependencies. Models trained on this data demonstrate significant improvements on RULER benchmarks, particularly in tasks requiring distant information. Following instruction fine-tuning, our models also achieve substantial gains on LongBenchv2, demonstrating enhanced long-context understanding. Extensive ablation studies further validate the necessity and effectiveness of entropybased verification for long-context training.
format Preprint
id arxiv_https___arxiv_org_abs_2510_02330
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle EntropyLong: Effective Long-Context Training via Predictive Uncertainty
Jia, Junlong
Chen, Ziyang
Wu, Xing
Gao, Chaochen
Lin, Zijia
Zhang, Debing
Hu, Songlin
Guo, Binghui
Computation and Language
Artificial Intelligence
Training long-context language models to capture long-range dependencies requires specialized data construction. Current approaches, such as generic text concatenation or heuristic-based variants, frequently fail to guarantee genuine long-range dependencies. We propose EntropyLong, a novel data construction method that leverages predictive uncertainty to verify dependency quality. Our approach identifies high-entropy positions in documents, retrieves semantically relevant contexts from large corpora, and verifies their utility by assessing whether they reduce prediction entropy. This model-in-the-loop verification ensures each dependency represents measurable information gain rather than spurious correlation. We construct training samples with long-range dependencies by combining original documents with these verified contextual supplements. Using FineWebEdu and Cosmopedia, we generate a dataset of 128K-length sequences with verified dependencies. Models trained on this data demonstrate significant improvements on RULER benchmarks, particularly in tasks requiring distant information. Following instruction fine-tuning, our models also achieve substantial gains on LongBenchv2, demonstrating enhanced long-context understanding. Extensive ablation studies further validate the necessity and effectiveness of entropybased verification for long-context training.
title EntropyLong: Effective Long-Context Training via Predictive Uncertainty
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2510.02330