The Common Pile v0.1: An 8TB Dataset of Public Domain and Openly Licensed Text

TL;DR

This paper introduces the Common Pile v0.1, an 8TB openly licensed text dataset for training large language models, demonstrating its effectiveness by training competitive 7-billion-parameter models.

Contribution

The authors curated and released a large, high-quality openly licensed dataset and validated its utility by training competitive LLMs, addressing data size and quality limitations of prior efforts.

Findings

Models trained on the Common Pile achieve performance comparable to models trained on unlicensed data.

The dataset covers diverse domains including research, code, books, and transcripts.

The authors release both the dataset and training code for community use.

Abstract

Large language models (LLMs) are typically trained on enormous quantities of unlicensed text, a practice that has led to scrutiny due to possible intellectual property infringement and ethical concerns. Training LLMs on openly licensed text presents a first step towards addressing these issues, but prior data collection efforts have yielded datasets too small or low-quality to produce performant LLMs. To address this gap, we collect, curate, and release the Common Pile v0.1, an eight terabyte collection of openly licensed text designed for LLM pretraining. The Common Pile comprises content from 30 sources that span diverse domains including research papers, code, books, encyclopedias, educational materials, audio transcripts, and more. Crucially, we validate our efforts by training two 7 billion parameter LLMs on text from the Common Pile: Comma v0.1-1T and Comma v0.1-2T, trained on 1 and 2 trillion tokens respectively. Both models attain competitive performance to LLMs trained on unlicensed text with similar computational budgets, such as Llama 1 and 2 7B. In addition to releasing the Common Pile v0.1 itself, we also release the code used in its creation as well as the training mixture and checkpoints for the Comma v0.1 models.