Saved in:
Bibliographic Details
Main Authors: Held, William, Paranjape, Bhargavi, Koura, Punit Singh, Lewis, Mike, Zhang, Frank, Mihaylov, Todor
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2501.11747
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916581306531840
author Held, William
Paranjape, Bhargavi
Koura, Punit Singh
Lewis, Mike
Zhang, Frank
Mihaylov, Todor
author_facet Held, William
Paranjape, Bhargavi
Koura, Punit Singh
Lewis, Mike
Zhang, Frank
Mihaylov, Todor
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2501_11747
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Optimizing Pretraining Data Mixtures with LLM-Estimated Utility
Held, William
Paranjape, Bhargavi
Koura, Punit Singh
Lewis, Mike
Zhang, Frank
Mihaylov, Todor
Computation and Language
Artificial Intelligence
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.
title Optimizing Pretraining Data Mixtures with LLM-Estimated Utility
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2501.11747