ACL Findings2023
Scaling Laws for BERT in Low-Resource Settings
Gorka Urbizu, Iñaki San Vicente, Xabier Saralegi, Rodrigo Agerri, Aitor Soroa
摘要
Large language models are very resource intensive, both financially and environmentally, and require an amount of training data which is simply unobtainable for the majority of NLP practitioners. Previous work has researched the scaling laws of such models, but optimal ratios of model parameters, dataset size, and computation costs focused on the large scale. In contrast, we analyze the effect those variables have on the performance of language models in constrained settings, by building three lightweight BERT models (16M/51M/124M parameters) trained over a set of small corpora (5M/25M/125M words). We experiment on four languages of different linguistic characteristics (Basque, Spanish, Swahili and Finnish), and evaluate the models on MLM and several NLU tasks. We conclude that the power laws for parameters, data and compute for lowresource settings differ from the optimal scaling laws previously inferred, and data requirements should be higher. Our insights are consistent across all the languages we study, as well as across the MLM and downstream tasks. Furthermore, we experimentally establish when the cost of using a Transformer-based approach is worth taking, instead of favouring other computationally lighter solutions.
问 Lune
几分钟读透这篇 Paper,不用再花一下午。
Lune 能读完全文,也能读引用它的顶会 Paper,讲清它的结论、证据和局限,每一点都注明出自哪篇 Paper。不用硬啃,也不靠猜。
已有 3 篇顶会 Paper 引用了它。别等审稿人问起,先弄清它们发现了什么。
脉络
- Unified Scaling Laws for Routed Language Models 引用了 Language Models are Few-Shot Learners
- Training compute-optimal transformer encoder models 引用了 Unsupervised Cross-lingual Representation Learning at Scale
- Inference Scaling Law for Retrieval Augmented Generation 引用了 Unsupervised Cross-lingual Representation Learning at Scale
- Inference Scaling Law for Retrieval Augmented Generation 引用了 Language Models are Few-Shot Learners
引用它的顶会 Paper3
- Training compute-optimal transformer encoder modelsMegi Dervishi, Alexandre Allauzen, Gabriel Synnaeve, Yann LeCunEMNLP 2025 · 1 次引用
- Inference Scaling Law for Retrieval Augmented GenerationShu Zhou, Yuxuan Ao, Yunyang Xuan, Xin Wang 等AAAI 2026 · 1 次引用
- From Retrieval to Translation: Translating Query into Graph-level Clues for Retrieval-Augmented GenerationQichuan Liu, Qinggang Zhang, Yuxuan Hu, Chenfeng Zheng 等ICML 2026
它引用的 Paper8
- Language Models are Few-Shot LearnersTom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah 等NeurIPS 2020 · 64,255 次引用
- Unsupervised Cross-lingual Representation Learning at ScaleAlexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary 等ACL 2020 · 539 次引用
- On the Stability of Fine-tuning BERT: Misconceptions, Explanations, and Strong BaselinesMarius Mosbach, Maksym Andriushchenko, Dietrich KlakowICLR 2021 · 448 次引用
- Unified Scaling Laws for Routed Language ModelsAidan Clark, Diego de Las Casas, Aurelia Guy, Arthur Mensch 等ICML 2022 · 266 次引用
- A Systematic Assessment of Syntactic Generalization in Neural Language ModelsJennifer Hu, Jon Gauthier, Peng Qian, Ethan Wilcox 等ACL 2020 · 124 次引用
相关 Paper
- Not Enough Data to Pre-train Your Language Model? MT to the Rescue!Gorka Urbizu, Iñaki San Vicente, Xabier Saralegi, Ander CorralACL Findings 2023 · 4 次引用
- From Zero to Hero: On the Limitations of Zero-Shot Language Transfer with Multilingual TransformersAnne Lauscher, Vinit Ravishankar, Ivan Vulic, Goran GlavasEMNLP 2020 · 235 次引用
- Language models scale reliably with over-training and on downstream tasksSamir Yitzhak Gadre, Georgios Smyrnis, Vaishaal Shankar, Suchin Gururangan 等ICLR 2025 · 3 次引用
- OWLS: Scaling Laws for Multilingual Speech Recognition and Translation ModelsWilliam Chen, Jinchuan Tian, Yifan Peng, Brian Yan 等ICML 2025
- Beyond Chinchilla-Optimal: Accounting for Inference in Language Model Scaling LawsNikhil Sardana, Jacob P. Portes, Sasha Doubov, Jonathan FrankleICML 2024 · 144 次引用
