Nemotron-CC: Transforming Common Crawl into a Refined Long-Horizon Pretraining Dataset

TL;DR

This paper introduces Nemotron-CC, a large, high-quality Common Crawl dataset optimized for long-horizon language model training, achieving state-of-the-art results with 15 trillion tokens and improved benchmark performance.

Contribution

It presents novel data filtering and augmentation techniques to create a large, high-quality dataset suitable for long-token horizon training, outperforming prior datasets like DCLM.

Findings

Improved MMLU scores by 5.6 with 1T tokens using high-quality subset.

Full 6.3T dataset matches DCLM on MMLU but with four times more unique tokens.

Training on 15T tokens with Nemotron-CC surpasses Llama 3.1 8B model on multiple benchmarks.

Abstract

Recent English Common Crawl datasets like FineWeb-Edu and DCLM achieved significant benchmark gains via aggressive model-based filtering, but at the cost of removing 90% of data. This limits their suitability for long token horizon training, such as 15T tokens for Llama 3.1. In this paper, we show how to achieve better trade-offs between accuracy and data quantity by a combination of classifier ensembling, synthetic data rephrasing, and reduced reliance on heuristic filters. When training 8B parameter models for 1T tokens, using a high-quality subset of our data improves MMLU by 5.6 over DCLM, demonstrating the efficacy of our methods for boosting accuracies over a relatively short token horizon. Furthermore, our full 6.3T token dataset matches DCLM on MMLU, but contains four times more unique real tokens than DCLM. This unlocks state-of-the-art training over a long token horizon: an 8B parameter model trained for 15T tokens, of which 7.2T came from our dataset, is better than the Llama 3.1 8B model: +5 on MMLU, +3.1 on ARC-Challenge, and +0.5 on average across ten diverse tasks. The dataset is available at https://data.commoncrawl.org/contrib/Nemotron/Nemotron-CC/index.html