Elamur Research⌕
☰
home / papers / 2025 / MegaMath: Pushing the Limits of Open Math Corpora

MegaMath: Pushing the Limits of Open Math Corpora

Paper · 2025 · Fan Zhou, Zengzhi Wang, Nikhil Ranjan et al.

Summary

Open math-focused LLMs have fallen behind proprietary systems, partly because no public pre-training corpus matches the size and quality of in-house collections such as DeepSeekMath's. MegaMath, from MBZUAI/LLM360, addresses this with a 371.6B-token English math corpus built from three sources. For web data, the authors reprocessed raw WARC files from Common Crawl. They rewrote math markup into LaTeX before extraction and ran a coarse-to-fine pipeline: Resiliparse with a fastText filter first, then deduplication, and finally trafilatura re-extraction with a retrained classifier. For code, a small fine-tuned model selected math-relevant snippets from Stack-V2. For synthetic data, they generated QA pairs, translated code into Python, and built mixed text-and-code samples. Every design choice was tested through ablation pre-training runs. Under equal token budgets, the curated subsets match or beat existing open math corpora. Continued training of Llama-3.2 models on the data brings sizeable gains in math reasoning. The fully released data and pipeline give open math model research a reproducible foundation.

Key points

  • Composition: roughly 279B tokens of web text (MegaMath-Web), 28.1B of code (MegaMath-Code), and 64.5B of synthetic data (MegaMath-Synthetic). The source covers 99 Common Crawl snapshots from 2014-15 through 2024-46. The dataset is available at hf.co/datasets/LLM360/MegaMath.
  • Classifier evaluation: the fastText math classifier was tested on an out-of-distribution suite drawn from MathPile (arXiv, StackExchange, Wikipedia, textbooks). Better text normalization and seed data lifted its average F1 from 81.8% to 98.8%. Deduplication used MinHash LSH with r=11, b=10 and a threshold of 0.75.
  • Premium subset: MegaMath-Web-Pro was produced with LM-based filtering and LLM refinement. In a 55B-token comparison, it scored at least 4% higher than FineMath-3+ and FineMath-4+.
  • Synthetic QA ablation: for QA generation, prompts in an ELI5 style that also stressed information completeness gave the best results (Core 48.8 vs. 34.6 for WebInstruct). Interleaved text-and-code blocks clearly beat raw code on both CoT and PAL.
  • Llama-3.2 training: Llama-3.2-1B was trained on 100B tokens and Llama-3.2-3B on 50B tokens, improving CoT scores by about 15–20%. The 3B model reached 56.2% on GSM8K and 25.1% on MATH.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 19,503 worksArtificial Intelligence · 18,805 worksArts and Humanities · 17,959 worksLanguage and Linguistics · 17,789 worksComputational Theory and Mathematics · 496 works

Cite

Cite as Fan Zhou, Zengzhi Wang, Nikhil Ranjan, Zhoujun Cheng, Liping Tang, Guowei He, Zhengzhong Liu and Eric P. Xing. 2025. MegaMath: Pushing the Limits of Open Math Corpora. arXiv:2504.02807.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2504.02807 · arXiv 2504.02807

@misc{zhou2025megamath,
  title = {{MegaMath: Pushing the Limits of Open Math Corpora}},
  author = {Fan Zhou, Zengzhi Wang, Nikhil Ranjan, Zhoujun Cheng, Liping Tang, Guowei He, Zhengzhong Liu and Eric P. Xing},
  year = {2025},
  doi = {10.48550/arxiv.2504.02807},
  eprint = {2504.02807},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2504.02807}
}

References

57 references, 33 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Marah Abdin, Jyoti Aneja, Hany Awadalla, Ahmed Awadallah, Ammar Ahmad Awan, Nguyen Bach, Amit Bahree, Arash Bakhtiari, Jianmin Bao, Harkirat Behl, et al. Phi-3 technical report: A highly capable language model locally on your phone. arXiv preprint arXiv:2404.14219, 2024.
  2. Loubna Ben Allal, Anton Lozhkov, Elie Bakouch, Gabriel Martı́n Blázquez, Guilherme Penedo, Lewis Tunstall, Andrés Marafioti, Hynek Kydlı́ček, Agustı́n Piqueres Lajarı́n, Vaibhav Srivastav, Joshua Lochner, Caleb Fahlgren, Xuan-Son Nguyen, Clémentine Fourrier, Ben Burtenshaw, Hugo Larcher, Haojun… — SmolLM2: When Smol Goes Big — Data-Centric Training of a Small Language Model
  3. Aida Amini, Saadia Gabriel, Shanchuan Lin, Rik Koncel-Kedziorski, Yejin Choi, and Hannaneh Hajishirzi. MathQA: Towards interpretable math word problem solving with operation-based formalisms. In Jill Burstein, Christy Doran, and Thamar Solorio (eds.), Proceedings of the 2019 Conference of the North…
  4. Viraat Aryabumi, Yixuan Su, Raymond Ma, Adrien Morisot, Ivan Zhang, Acyr Locatelli, Marzieh Fadaee, Ahmet Üstün, and Sara Hooker. To code, or not to code? exploring impact of code in pre-training. arXiv preprint arXiv:2408.10914, 2024. — [arXiv:2408.10914](https://arxiv.org/abs/2408.10914) find
  5. Zhangir Azerbayev, Hailey Schoelkopf, Keiran Paster, Marco Dos Santos, Stephen McAleer, Albert Q Jiang, Jia Deng, Stella Biderman, and Sean Welleck. Llemma: An open language model for mathematics. arXiv preprint arXiv:2310.10631, 2023.
  6. Piotr Bojanowski, Edouard Grave, Armand Joulin, and Tomas Mikolov. Enriching word vectors with subword information. Transactions of the association for computational linguistics, 5:135–146, 2017.
  7. Andrei Z Broder. On the resemblance and containment of documents. In Proceedings. Compression and Complexity of SEQUENCES 1997 (Cat. No. 97TB100171), pp. 21–29. find
  8. Andrei Z Broder. Identifying and filtering near-duplicate documents. In Annual symposium on combinatorial pattern matching, pp. 1–10. Springer, 2000. find
  9. Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, et al. Training verifiers to solve math word problems. arXiv preprint arXiv:2110.14168, 2021.
  10. Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, et al. The llama 3 herd of models. arXiv preprint arXiv:2407.21783, 2024.
  11. Luyu Gao, Aman Madaan, Shuyan Zhou, Uri Alon, Pengfei Liu, Yiming Yang, Jamie Callan, and Graham Neubig. Pal: Program-aided language models. In International Conference on Machine Learning, pp. 10764–10799. PMLR, 2023. find
  12. Zhibin Gou, Zhihong Shao, Yeyun Gong, Yujiu Yang, Minlie Huang, Nan Duan, Weizhu Chen, et al. Tora: A tool-integrated reasoning agent for mathematical problem solving. In The Twelfth International Conference on Learning Representations, 2024.
  13. Suriya Gunasekar, Yi Zhang, Jyoti Aneja, Caio César Teodoro Mendes, Allie Del Giorno, Sivakanth Gopi, Mojan Javaheripi, Piero Kauffmann, Gustavo de Rosa, Olli Saarikivi, et al. Textbooks are all you need. arXiv preprint arXiv:2306.11644, 2023. find
  14. Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. arXiv preprint arXiv:2501.12948, 2025.
  15. Xiaotian Han, Yiren Jian, Xuefeng Hu, Haogeng Liu, Yiqi Wang, Qihang Fan, Yuang Ai, Huaibo Huang, Ran He, Zhenheng Yang, et al. Infimm-webmath-40b: Advancing multimodal pre-training for enhanced mathematical reasoning. In The 4th Workshop on Mathematical Reasoning and AI at NeurIPS’24, 2024. find
  16. Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. Measuring massive multitask language understanding. arXiv preprint arXiv:2009.03300, 2020.
  17. Dan Hendrycks, Collin Burns, Saurav Kadavath, Akul Arora, Steven Basart, Eric Tang, Dawn Song, and Jacob Steinhardt. Measuring mathematical problem solving with the math dataset. In Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 2), 2021.
  18. Shengding Hu, Yuge Tu, Xu Han, Chaoqun He, Ganqu Cui, Xiang Long, Zhi Zheng, Yewei Fang, Yuxiang Huang, Weilin Zhao, et al. Minicpm: Unveiling the potential of small language models with scalable training strategies. arXiv preprint arXiv:2404.06395, 2024.
  19. Binyuan Hui, Jian Yang, Zeyu Cui, Jiaxi Yang, Dayiheng Liu, Lei Zhang, Tianyu Liu, Jiajun Zhang, Bowen Yu, Keming Lu, et al. Qwen2. 5-coder technical report. arXiv preprint arXiv:2409.12186, 2024.
  20. Armand Joulin, Edouard Grave, Piotr Bojanowski, and Tomas Mikolov. Bag of tricks for efficient text classification. arXiv preprint arXiv:1607.01759, 2016. find
  21. Rik Koncel-Kedziorski, Subhro Roy, Aida Amini, Nate Kushman, and Hannaneh Hajishirzi. MAWPS: A math word problem repository. In Kevin Knight, Ani Nenkova, and Owen Rambow (eds.), Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human… find
  22. Katherine Lee, Daphne Ippolito, Andrew Nystrom, Chiyuan Zhang, Douglas Eck, Chris Callison-Burch, and Nicholas Carlini. Deduplicating training data makes language models better. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics. Association for Computational… find
  23. Aitor Lewkowycz, Anders Andreassen, David Dohan, Ethan Dyer, Henryk Michalewski, Vinay Ramasesh, Ambrose Slone, Cem Anil, Imanol Schlag, Theo Gutman-Solo, et al. Solving quantitative reasoning problems with language models. Advances in Neural Information Processing Systems, 35:3843–3857, 2022.
  24. Jeffrey Li, Alex Fang, Georgios Smyrnis, Maor Ivgi, Matt Jordan, Samir Gadre, Hritik Bansal, Etash Guha, Sedrick Keh, Kushal Arora, Saurabh Garg, Rui Xin, Niklas Muennighoff, Reinhard Heckel, Jean Mercat, Mayee Chen, Suchin Gururangan, Mitchell Wortsman, Alon Albalak, Yonatan Bitton, Marianna Nezhur… — DataComp-LM: In search of the next generation of training sets for language models
  25. Jia Li, Edward Beeching, Lewis Tunstall, Ben Lipkin, Roman Soletskyi, Shengyi Huang, Kashif Rasul, Longhui Yu, Albert Q Jiang, Ziju Shen, et al. Numinamath: The largest public dataset in ai4maths with 860k pairs of competition math problems and solutions. Hugging Face repository, 13, 2024b. find
  26. Aixin Liu, Bei Feng, Bing Xue, Bingxuan Wang, Bochao Wu, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, et al. Deepseek-v3 technical report. arXiv preprint arXiv:2412.19437, 2024. find
  27. Anton Lozhkov, Loubna Ben Allal, Elie Bakouch, Leandro von Werra, and Thomas Wolf. Finemath: the finest collection of mathematical content, 2024a. URL https://huggingface.co/datasets/HuggingFaceTB/finemath. ↗
  28. Anton Lozhkov, Raymond Li, Loubna Ben Allal, Federico Cassano, Joel Lamy-Poirier, Nouamane Tazi, Ao Tang, Dmytro Pykhtar, Jiawei Liu, Yuxiang Wei, et al. Starcoder 2 and the stack v2: The next generation. arXiv preprint arXiv:2402.19173, 2024b. find
  29. Pan Lu, Liang Qiu, Kai-Wei Chang, Ying Nian Wu, Song-Chun Zhu, Tanmay Rajpurohit, Peter Clark, and Ashwin Kalyan. Dynamic prompt learning via policy gradient for semi-structured mathematical reasoning. In The Eleventh International Conference on Learning Representations, 2023.
  30. Zimu Lu, Aojun Zhou, Ke Wang, Houxing Ren, Weikang Shi, Junting Pan, Mingjie Zhan, and Hongsheng Li. Mathcoder2: Better math reasoning from continued pretraining on model-translated mathematical code. arXiv preprint arXiv:2410.08196, 2024.
  31. Pratyush Maini, Skyler Seto, Richard Bai, David Grangier, Yizhe Zhang, and Navdeep Jaitly. Rephrasing the web: A recipe for compute and data-efficient language modeling. In Lun-Wei Ku, Andre Martins, and Vivek Srikumar (eds.), Proceedings of the 62nd Annual Meeting of the Association for Computation…
  32. Shen-yun Miao, Chao-Chun Liang, and Keh-Yih Su. A diverse corpus for evaluating and developing English math word problem solvers. In Dan Jurafsky, Joyce Chai, Natalie Schluter, and Joel Tetreault (eds.), Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pp. 975… find
  33. OpenAI. Learning to reason with llms, 2024. learning-to-reason-with-llms/. find
  34. Keiran Paster, Marco Dos Santos, Zhangir Azerbayev, and Jimmy Ba. Openwebmath: An open dataset of high-quality mathematical web text. In The Twelfth International Conference on Learning Representations, 2024.
  35. Arkil Patel, Satwik Bhattamishra, and Navin Goyal. Are NLP models really able to solve simple math word problems? In Kristina Toutanova, Anna Rumshisky, Luke Zettlemoyer, Dilek Hakkani-Tur, Iz Beltagy, Steven Bethard, Ryan Cotterell, Tanmoy Chakraborty, and Yichao Zhou (eds.), Proceedings of the 202…
  36. Guilherme Penedo, Quentin Malartic, Daniel Hesslow, Ruxandra Cojocaru, Alessandro Cappelli, Hamza Alobeidli, Baptiste Pannier, Ebtesam Almazrouei, and Julien Launay. The refinedweb dataset for falcon llm: outperforming curated corpora with web data, and web data only. arXiv preprint arXiv:2306.01116… find
  37. Guilherme Penedo, Hynek Kydlı́ček, Anton Lozhkov, Margaret Mitchell, Colin Raffel, Leandro Von Werra, Thomas Wolf, et al. The fineweb datasets: Decanting the web for the finest text data at scale. arXiv preprint arXiv:2406.17557, 2024.
  38. Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified text-to-text transformer. Journal of machine learning research, 21(140):1–67, 2020.
  39. Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Y Wu, et al. Deepseekmath: Pushing the limits of mathematical reasoning in open language models. arXiv preprint arXiv:2402.03300, 2024.
  40. URL — Nemotron-CC: Transforming Common Crawl into a Refined Long-Horizon Pretraining Dataset https://openai.com/index/Dan Su, Kezhi Kong, Ying Lin, Joseph Jennings, Brandon Norick, Markus Kliegl, Mostofa Patwary, Mohammad Shoeybi, and Bryan Catanzaro. Nemotron-cc: Transforming common crawl into a refined long-horizon pretraining dataset. arXiv preprint arXiv:2412.02595, 2024. ↗
  41. Liping Tang, Nikhil Ranjan, Omkar Pangarkar, Xuezhi Liang, Zhen Wang, Li An, Bhaskar Rao, Linghao Jin, Huijuan Wang, Zhoujun Cheng, Suqi Sun, Cun Mu, Victor Miller, Xuezhe Ma, Yue Peng, Zhengzhong Liu, and Eric P. Xing. Txt360: A top-quality llm pre-training dataset requires the perfect blend, 2024.… find
  42. Yury Tokpanov, Paolo Glorioso, Quentin Anthony, and Beren Millidge. Zyda-2: a 5 trillion token high-quality dataset. arXiv preprint arXiv:2411.06068, 2024. — [arXiv:2411.06068](https://arxiv.org/abs/2411.06068) find
  43. Shubham Toshniwal, Wei Du, Ivan Moshkov, Branislav Kisacanin, Alexan Ayrapetyan, and Igor Gitman. Openmathinstruct-2: Accelerating ai for math with massive open-source instruction data. arXiv preprint arXiv:2410.01560, 2024.
  44. Zengzhi Wang, Xuefeng Li, Rui Xia, and Pengfei Liu. Mathpile: A billion-token-scale pretraining corpus for math. In A. Globerson, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. Tomczak, and C. Zhang (eds.), Advances in Neural Information Processing Systems, volume 37, pp. 25426–25468. Curran Associat…
  45. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35:24824–24837, 2022.
  46. Yuxiang Wei, Hojae Han, and Rajhans Samdani. Arctic-snowcoder: Demystifying high-quality data in code pretraining. arXiv preprint arXiv:2409.02326, 2024. — [arXiv:2409.02326](https://arxiv.org/abs/2409.02326) find
  47. Ruijie Xu, Zengzhi Wang, Run-Ze Fan, and Pengfei Liu. Benchmarking benchmark leakage in large language models. arXiv preprint arXiv:2404.18824, 2024a.
  48. Yiheng Xu, Hongjin Su, Chen Xing, Boyu Mi, Qian Liu, Weijia Shi, Binyuan Hui, Fan Zhou, Yitao Liu, Tianbao Xie, Zhoujun Cheng, Siheng Zhao, Lingpeng Kong, Bailin Wang, Caiming Xiong, and Tao Yu. Lemur: Harmonizing natural language and code for language agents. In The Twelfth International Conference… find
  49. OpenReview.net, 2024b. URL https://openreview.net/forum?id=hNhwSmtXRh. ↗
  50. An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, et al. Qwen2.5 technical report. arXiv preprint arXiv:2412.15115, 2024a. find
  51. An Yang, Beichen Zhang, Binyuan Hui, Bofei Gao, Bowen Yu, Chengpeng Li, Dayiheng Liu, Jianhong Tu, Jingren Zhou, Junyang Lin, et al. Qwen2. 5-math technical report: Toward mathematical expert model via self-improvement. arXiv preprint arXiv:2409.12122, 2024b.
  52. Weizhe Yuan, Richard Yuanzhe Pang, Kyunghyun Cho, Sainbayar Sukhbaatar, Jing Xu, and Jason Weston. Self-rewarding language models. arXiv preprint arXiv:2401.10020, 2024. find
  53. Xiang Yue, Tuney Zheng, Ge Zhang, and Wenhu Chen. Mammoth2: Scaling instructions from the web. arXiv preprint arXiv:2405.03548, 2024.
  54. Liang Zeng, Liangjun Zhong, Liang Zhao, Tianwen Wei, Liu Yang, Jujie He, Cheng Cheng, Rui Hu, Yang Liu, Shuicheng Yan, et al. Skywork-math: Data scaling laws for mathematical reasoning in large language models–the story goes on. arXiv preprint arXiv:2407.08348, 2024.
  55. Peiyuan Zhang, Guangtao Zeng, Tianduo Wang, and Wei Lu. Tinyllama: An open-source small language model. arXiv preprint arXiv:2401.02385, 2024.
  56. Yaowei Zheng, Richong Zhang, Junhao Zhang, Yanhan Ye, Zheyan Luo, Zhangchi Feng, and Yongqiang Ma. Llamafactory: Unified efficient fine-tuning of 100+ language models. arXiv preprint arXiv:2403.13372, 2024.
  57. Fan Zhou, Zengzhi Wang, Qian Liu, Junlong Li, and Pengfei Liu. Programming every example: Lifting pre-training data quality like experts at scale. arXiv preprint arXiv:2409.17115, 2024. — [arXiv:2409.17115](https://arxiv.org/abs/2409.17115) find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhou-2025-megamath-pushing-limits-open-math/