Research map / Large language models / Agents, code and math
Nemotron-CC-Math: A 133 Billion-Token-Scale High Quality Math Pretraining Dataset
Rabeeh Karimi mahabadi, Sanjeev Satheesh, Shrimai Prabhumoye, Mostofa Patwary, Mohammad Shoeybi, Bryan Catanzaro
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| pretrain on Nemotron-CC-Math-4+ instead of FineMath-4+ | Nemotron-T 8B · MATH (100B tokens) | EM: 35.80 -> 40.60 | Improved |
| pretrain on Nemotron-CC-Math-4+ instead of MegaMath-Pro | Nemotron-T 8B · MATH (100B tokens) | EM: 34.00 -> 40.60 | Improved |
| pretrain on Nemotron-CC-Math-4+ instead of FineMath-4+ | Nemotron-T 8B · GSM8K (100B tokens) | EM: 75.97 -> 76.27 | Improved |
| pretrain on Nemotron-CC-Math-4+ instead of MegaMath-Pro | Nemotron-T 8B · MBPP+ (100B tokens) | avg@20: 46.03 -> 45.11 | Hurt |
| pretrain on Nemotron-CC-Math-4+ instead of FineMath-4+ | Nemotron-T 8B · MBPP+ (100B tokens) | avg@20: 28.88 -> 45.11 | Improved |
| pretrain on Nemotron-CC-Math-3+ instead of FineMath-3+ | Nemotron-T 8B · MATH (300B tokens) | EM: 34.60 -> 44.20 | Improved |
| pretrain on Nemotron-CC-Math-3+ instead of MegaMath-Web | Nemotron-T 8B · MATH (300B tokens) | EM: 31.60 -> 44.20 | Improved |
| pretrain on Nemotron-CC-Math-3+ instead of FineMath-3+ | Nemotron-T 8B · MBPP+ (300B tokens) | avg@20: 29.19 -> 43.51 | Improved |
| pretrain on Nemotron-CC-Math-3+ instead of FineMath-3+ | Nemotron-T 8B · MMLU-Pro (300B tokens) | EM: 39.57 -> 39.32 | No clear effect |
| pretrain on Nemotron-CC-Math-3+ instead of FineMath-3+ | Nemotron-T 8B · MBPP (300B tokens) | avg@20: 57.57 -> 56.15 | Hurt |
| pretrain on Nemotron-CC-Math-3+ instead of FineMath-3+ | Nemotron-T 8B · MMLU-Stem (300B tokens) | EM: 62.29 -> 64.26 | Improved |
| use Phi-4 (14B) instead of DeepSeek-V3 (671B) for boilerplate cleaning (ablation) | cleaning pipeline · GSM8K | EM: 75.51 -> 79.98 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- OpenWebMath — OpenWebMath: An Open Dataset of High-Quality Mathematical Web Text (ICLR 2024) · better on 8 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- OpenWebMath: An Open Dataset of High-Quality Mathematical Web Text (ICLR 2024 · similarity 0.91)
- MathPile: A Billion-Token-Scale Pretraining Corpus for Math (NeurIPS 2024 · similarity 0.90)
- MathCoder2: Better Math Reasoning from Continued Pretraining on Model-translated Mathematical Code (ICLR 2025 · similarity 0.89)
- OpenMathInstruct-2: Accelerating AI for Math with Massive Open-Source Instruction Data (ICLR 2025 · similarity 0.89)
- OpenMathInstruct-1: A 1.8 Million Math Instruction Tuning Dataset (NeurIPS 2024 · similarity 0.88)
- MASS: Mathematical Data Selection via Skill Graphs for Pretraining Large Language Models (ICML 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
