Does your data spark joy? Performance gains from domain upsampling at the end of training

TL;DR

This paper demonstrates that end-of-training domain upsampling of smaller datasets significantly improves performance on difficult benchmarks for large language models, offering a cost-effective way to optimize pretraining data composition.

Contribution

It introduces a simple end-of-training upsampling technique for domain-specific data that enhances model performance and allows scalable dataset utility analysis.

Findings

Upsampling domain data at training end improves benchmark scores by up to 8.26 pp.

Optimal upsampling proportion is between 10% and 20%.

The method rivals longer training models with less computational cost.

Abstract

Pretraining datasets for large language models (LLMs) have grown to trillions of tokens composed of large amounts of CommonCrawl (CC) web scrape along with smaller, domain-specific datasets. It is expensive to understand the impact of these domain-specific datasets on model capabilities as training at large FLOP scales is required to reveal significant changes to difficult and emergent benchmarks. Given the increasing cost of experimenting with pretraining data, how does one determine the optimal balance between the diversity in general web scrapes and the information density of domain specific data? In this work, we show how to leverage the smaller domain specific datasets by upsampling them relative to CC at the end of training to drive performance improvements on difficult benchmarks. This simple technique allows us to improve up to 6.90 pp on MMLU, 8.26 pp on GSM8K, and 6.17 pp on HumanEval relative to the base data mix for a 7B model trained for 1 trillion (T) tokens, thus rivaling Llama-2 (7B)$\unicode{x2014}$a model trained for twice as long. We experiment with ablating the duration of domain upsampling from 5% to 30% of training and find that 10% to 20% percent is optimal for navigating the tradeoff between general language modeling capabilities and targeted benchmarks. We also use domain upsampling to characterize at scale the utility of individual datasets for improving various benchmarks by removing them during this final phase of training. This tool opens up the ability to experiment with the impact of different pretraining datasets at scale, but at an order of magnitude lower cost compared to full pretraining runs.