paper-with-me

홈 › Papers

Optimizing Pretraining Data Mixtures with LLM-Estimated Utility

2025-01-20 · William Held, Bhargavi Paranjape, Punit Singh Koura, Mike Lewis, Frank Zhang, Todor Mihaylov

Large Language Models improve with increasing amounts of high-quality training data. However, leveraging larger datasets requires balancing quality, quantity, and diversity across sources. After evaluating nine baseline methods under both compute- and data-constrained scenarios, we find token-count heuristics outperform manual and learned mixes, indicating that simple approaches accounting for dataset size and diversity are surprisingly effective. Building on this insight, we propose two complementary approaches: UtiliMax, which extends token-based heuristics by incorporating utility estimates from reduced-scale ablations, achieving up to a 10.6x speedup over manual baselines; and Model Estimated Data Utility (MEDU), which leverages LLMs to estimate data utility from small samples, matching ablation-based performance while reducing computational requirements by $\sim$200x. Together, these approaches establish a new framework for automated, compute-efficient data mixing that is robust across training regimes.

📄 PDF Abstract BibTeX arXiv:2501.11747

Code (0)

등록된 구현이 없습니다.

Tasks

Diversity

Similar Papers 제목 키워드 기반

BiMix: A Bivariate Data Mixing Law for Language Model Pretraining

2024-05-23 · Ce Ge, Zhijian Ma, Daoyuan Chen, Yaliang Li 외

Large language models have demonstrated remarkable capabilities across various tasks, primarily attributed to the utilization of diversely sourced data. However, the impact of pretraining data composition on model perfor…

Language ModelingLanguage Modelling

Scaling Laws for Optimal Data Mixtures

2025-07-12 · Mustafa Shukor, Louis Bethune, Dan Busbridge, David Grangier 외 arxiv

Large foundation models are typically trained on data from multiple domains, with the data mixture--the proportion of each domain used--playing a critical role in model performance. The standard approach to selecting thi…

Data Mixing Laws: Optimizing Data Mixtures by Predicting Language Modeling Performance

2024-03-25 · Jiasheng Ye, Peiju Liu, Tianxiang Sun, Yunhua Zhou 외

Pretraining data of large language models composes multiple domains (e.g., web texts, academic papers, codes), whose mixture proportions crucially impact the competence of outcome models. While existing endeavors rely on…

Language ModelingLanguage Modelling

Bayesian Inference for Multidimensional Welfare Comparisons

2024-06-19 · David Gunawan, William Griffiths, Duangkamon Chotikapanich

Using both single-index measures and stochastic dominance concepts, we show how Bayesian inference can be used to make multivariate welfare comparisons. A four-dimensional distribution for the well-being attributes incom…

AttributeBayesian Inference

DoReMi: Optimizing Data Mixtures Speeds Up Language Model Pretraining

2023-05-17 · NeurIPS 2023 11 · Sang Michael Xie, Hieu Pham, Xuanyi Dong, Nan Du 외

The mixture proportions of pretraining data domains (e.g., Wikipedia, books, web text) greatly affect language model (LM) performance. In this paper, we propose Domain Reweighting with Minimax Optimization (DoReMi), whic…

Language ModelingLanguage Modelling