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.

Large language models have demonstrated remarkable capabilities across diverse tasks, yet their performance heavily relies on the quality and scale of pre-training data. While recent efforts have focused on improving data quality for English, most languages lack comparable resources. We introduce FineWeb2, a comprehensive pipeline for processing web data that adapts to any language while maintaining consistent quality standards. Our approach combines language-agnostic filtering techniques with language-specific optimizations, enabling the creation of high-quality pre-training datasets across 95 languages. We demonstrate that models trained on FineWeb2 data consistently outperform those trained on existing multilingual datasets, with particularly strong improvements for low-resource languages. By open-sourcing both our pipeline and processed datasets, we aim to democratize access to high-quality pre-training data and accelerate multilingual AI development.

While large language models have facilitated breakthroughs in many applications of artificial intelligence, their inherent largeness makes them computationally expensive and challenging to deploy in resource-constrained settings. In this paper, we document the development of SmolLM2, a state-of-the-art "small" (1.7 billion parameter) language model (LM). To attain strong performance, we overtrain SmolLM2 on ~11 trillion tokens of data using a multi-stage training process that mixes web text with specialized math, code, and instruction-following data. We additionally introduce new specialized datasets (FineMath, Stack-Edu, and SmolTalk) at stages where we found existing datasets to be problematically small or low-quality. To inform our design decisions, we perform both small-scale ablations as well as a manual refinement process that updates the dataset mixing rates at each stage based on the performance at the previous stage. Ultimately, we demonstrate that SmolLM2 outperforms other recent small LMs including Qwen2.5-1.5B and Llama3.2-1B. To facilitate future research on LM development as well as applications of small LMs, we release both SmolLM2 as well as all of the datasets we prepared in the course of this project.

Many AI companies are training their large language models (LLMs) on data without the permission of the copyright owners. The permissibility of doing so varies by jurisdiction: in countries like the EU and Japan, this is allowed under certain restrictions, while in the United States, the legal landscape is more ambiguous. Regardless of the legal status, concerns from creative producers have led to several high-profile copyright lawsuits, and the threat of litigation is commonly cited as a reason for the recent trend towards minimizing the information shared about training datasets by both corporate and public interest actors. This trend in limiting data information causes harm by hindering transparency, accountability, and innovation in the broader ecosystem by denying researchers, auditors, and impacted individuals access to the information needed to understand AI models. While this could be mitigated by training language models on open access and public domain data, at the time of writing, there are no such models (trained at a meaningful scale) due to the substantial technical and sociological challenges in assembling the necessary corpus. These challenges include incomplete and unreliable metadata, the cost and complexity of digitizing physical records, and the diverse set of legal and technical skills required to ensure relevance and responsibility in a quickly changing landscape. Building towards a future where AI systems can be trained on openly licensed data that is responsibly curated and governed requires collaboration across legal, technical, and policy domains, along with investments in metadata standards, digitization, and fostering a culture of openness.

The performance of a large language model (LLM) depends heavily on the quality and size of its pretraining dataset. However, the pretraining datasets for state-of-the-art open LLMs like Llama 3 and Mixtral are not publicly available and very little is known about how they were created. In this work, we introduce FineWeb, a 15-trillion token dataset derived from 96 Common Crawl snapshots that produces better-performing LLMs than other open pretraining datasets. To advance the understanding of how best to curate high-quality pretraining datasets, we carefully document and ablate all of the design choices used in FineWeb, including in-depth investigations of deduplication and filtering strategies. In addition, we introduce FineWeb-Edu, a 1.3-trillion token collection of educational text filtered from FineWeb. LLMs pretrained on FineWeb-Edu exhibit dramatically better performance on knowledge- and reasoning-intensive benchmarks like MMLU and ARC. Along with our datasets, we publicly release our data curation codebase and all of the models trained during our ablation experiments.

The performance of a large language model (LLM) depends heavily on the quality and size of its pretraining dataset. However, the pretraining datasets for state-of-the-art open LLMs like Llama 3 and Mixtral are not publicly available and very little is known about how they were created. In this work, we introduce FineWeb, a 15-trillion token dataset derived from 96 Common Crawl snapshots that produces better-performing LLMs than other open pretraining datasets. To advance the understanding of how best to curate high-quality pretraining datasets, we carefully document and ablate all of the design choices used in FineWeb, including in-depth investigations of deduplication and filtering strategies. In addition, we introduce FineWeb-Edu, a 1.3-trillion token collection of educational text filtered from FineWeb. LLMs pretrained on FineWeb-Edu exhibit dramatically better performance on knowledge- and reasoning-intensive benchmarks like MMLU and ARC. Along with our datasets, we publicly release our data curation codebase and all of the models trained during our ablation experiments.

We introduce the Falcon series: 7B, 40B, and 180B parameters causal decoder-only models trained on a diverse high-quality corpora predominantly assembled from web data. The largest model, Falcon-180B, has been trained on over 3.5 trillion tokens of text--the largest openly documented pretraining run. Falcon-180B significantly outperforms models such as PaLM or Chinchilla, and improves upon concurrently developed models such as LLaMA 2 or Inflection-1. It nears the performance of PaLM-2-Large at a reduced pretraining and inference cost, making it, to our knowledge, one of the three best language models in the world along with GPT-4 and PaLM-2-Large. We report detailed evaluations, as well as a deep dive into the methods and custom tooling employed to pretrain Falcon. Notably, we report on our custom distributed training codebase, allowing us to efficiently pretrain these models on up to 4,096 A100s on cloud AWS infrastructure with limited interconnect. We release a 600B tokens extract of our web dataset, as well as the Falcon-7/40/180B models under a permissive license to foster open-science and accelerate the development of an open ecosystem of large language models.

Large language models are commonly trained on a mixture of filtered web data and curated high-quality'' corpora, such as social media conversations, books, or technical papers. This curation process is believed to be necessary to produce performant models with broad zero-shot generalization abilities. However, as larger models requiring pretraining on trillions of tokens are considered, it is unclear how scalable is curation, and whether we will run out of unique high-quality data soon. At variance with previous beliefs, we show that properly filtered and deduplicated web data alone can lead to powerful models; even significantly outperforming models trained on The Pile. Despite extensive filtering, the high-quality data we extract from the web is still plentiful, and we are able to obtain five trillion tokens from CommonCrawl. We publicly release an extract of 500 billion tokens from our RefinedWeb dataset, and 1.3/7.5B parameters language models trained on it.

Recent advances in the space of Arabic large language models have opened up a wealth of potential practical applications. From optimal training strategies, large scale data acquisition and continuously increasing NLP resources, the Arabic LLM landscape has improved in a very short span of time, despite being plagued by training data scarcity and limited evaluation resources compared to English. In line with contributing towards this ever-growing field, we introduce AlGhafa, a new multiple-choice evaluation benchmark for Arabic LLMs. For showcasing purposes, we train a new suite of models, including a 14 billion parameter model, the largest monolingual Arabic decoder-only model to date. We use a collection of publicly available datasets, as well as a newly introduced HandMade dataset consisting of 8 billion tokens. Finally, we explore the quantitative and qualitative toxicity of several Arabic models, comparing our models to existing public Arabic LLMs.

The Artery in Microgravity (AIM) project was the first experiment to be selected for the “Orbit Your Thesis!” programme of the European Space Agency Academy. It is a 2U cube experiment that will be operated in the International Commercial Experiment (ICE) Cubes facility onboard the International Space Station. The experiment is expected to be launched on SpaceX-25 in mid-2022. The project is being developed by an international group of students from ISAE-SUPAERO and Politecnico di Torino. The objective of the experiment is to study haemodynamics in the space environment applied to coronary heart disease. The outcomes of this testbench will contribute to understanding the effects of radiation and microgravity on the circulatory system of an astronaut, specifically the behaviour in long-term human spaceflight. It will also help to ascertain the feasibility of individuals suffering from this kind of disease going to space someday. The cornerstones of the experiment are two models of 3D-printed artificial arteries, in stenotic and stented conditions respectively. Blood-mimicking fluid composed of water and glycerol is circulated through the arteries in a closed hydraulic loop, and a red dye is injected for flow visualisation. Drops of pressure and image analysis of the flow will be studied with the corresponding sensors and camera. The pH of the fluid will also be monitored to assess the effect of augmented radiation levels on the release of particles from the metallic stent. Some delays were experienced in the project due to the COVID-19 pandemic and to implement design improvements. Improvements were made to several aspects of the design including mechanics (e.g. remanufacturing the reservoir with surface treatment against corrosion, leak prevention measures), software (e.g. upgrading to Odroid-C4 and migrating the code to Python), and electronics (e.g. several iterations of the interface PCB design). This iterative process of identifying areas of concern and designing and implementing solutions has resulted in many lessons learned. The paper will outline in detail Phase D – Qualification and Production of the AIM experiment cube, with special insight on the implementation of the improvements. Previously, at the Symposium on Space Educational Activities in 2019 in Leicester, the initial phases of the design and development of the cube were presented. This year, the final flight model and the results of validation testing before launching on SpaceX-25 are presented. Lessons learned throughout the course of the project are also highlighted for students embarking on their own space-related educational activities.