datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
PolyMath
Dataset Card for PolyMath
Dataset Summary
PolyMath is a curated dataset of 11,090 high-difficulty mathematical problems designed for training reasoning models. Built for the AIMO Math Corpus Prize. Existing math datasets (NuminaMath-1.5, OpenMathReasoning) suffer from high noise rates in their hardest samples and largely unusable proof-based problems.
PolyMath addresses both issues through:
Data scraping: problems sourced from official competition PDFs absent from… See the full description on the dataset page: https://huggingface.co/datasets/AIMO-Corpus/PolyMath.AIMO3_CoT
AIMO3 CoT Dataset
数据集来源与目的 (Dataset Origin and Purpose)
本数据集源自 Kaggle 竞赛 AI Mathematical Olympiad - Progress Prize 3。
动机 (Motivation)
原始数据集仅包含问题和答案，缺乏思维链（Chain of Thought, CoT）。直接使用原始数据训练如 DeepSeek Math 或 Qwen Math 等模型效果不佳。因此，本项目的目的是利用 Gemini 3 Pro 为这些问题补充详细的 CoT，以提升模型在数学推理任务上的表现。
CoT 格式 (CoT Format)
生成的 CoT 遵循 ReAct 风格的推理过程，并使用中文叙述：
Thought: 分析问题并规划下一步。
Code: 编写 Python 代码进行计算或验证。
Observation: 代码的执行输出。
... (重复上述步骤)
Final Answer: 得出的最终答案。… See the full description on the dataset page: https://huggingface.co/datasets/UR-xiaoyang/AIMO3_CoT.Train_dataaimo-validation-aime-th
AIMO validation AIME-th
The 90 problems of AIME 2022, 2023 and 2024 — thirty each — with the problem statements
translated to Thai. Every row corresponds 1:1, in order, to a row of
AI-MO/aimo-validation-aime, and id, url and answer are identical to it.
Read this before scoring the solution column
38 of the 90 solutions are the English text, not Thai. The original translation
pass rendered every problem and skipped these solutions entirely. They are marked… See the full description on the dataset page: https://huggingface.co/datasets/iapp/aimo-validation-aime-th.QWQ-LongCOT-AIMOQWQ-LongCOT-AIMO is a derived dataset created by processing the amphora/QwQ-LongCoT-130K dataset. It filters the original dataset to focus specifically on question-answering pairs where the final answer is a numerical value between 0 and 999, explicitly marked using the \boxed{...} format within the original chain-of-thought answer.
Dataset Structure
Data Splits
The dataset is split into training, validation, and test sets with an 80/10/10 ratio based on the filtered… See the full description on the dataset page: https://huggingface.co/datasets/Floppanacci/QWQ-LongCOT-AIMO.aimo-dataset2500 challenging math QA data combining historical AIME, AMC12 and AMC10, and LIMO 900. Extracted and cleaned from billxbf/aimo-math-problems.
The hard version (1250) is filtered on failed pass@1 inference by Deekseek-Distill-Qwen-7B, with plain math COT.
aimo-3-eval
AIMO 3 Evaluation Dataset
This dataset is designed for evaluating mathematical reasoning models, specifically for the AIMO (AI Mathematical Olympiad) competition. It consolidates problems from multiple high-quality sources into a unified format.
Dataset Structure
The dataset contains the following columns:
id (string): A unique identifier for each problem, prefixed with the source name (e.g., aime-1, math500-test-123).
question (string): The text of the mathematical… See the full description on the dataset page: https://huggingface.co/datasets/abdullahmeda/aimo-3-eval.moh_8_rollouts_tool_calls
MOH 8 Rollouts Tool Calls
Fixed 8-rollout benchmark sampled from aimosprite/training-output using only tool-calling attempts.
Construction
Source problem family: polymath_*
Source eligibility band: correct_count_16 in [1, 10]
Candidate attempt pool per problem: attempts with Python Calls > 0
Final sample per problem: 8 attempts, sampled without replacement using seed 42
Additional constraint: the sampled 8 always include at least one correct attempt
Published rows: 300… See the full description on the dataset page: https://huggingface.co/datasets/aimosprite/moh_8_rollouts_tool_calls.AIMO-2_ReferenceThis CSV file is reference.csv in Kaggle's AI Mathematical Olympiad - Progress Prize 2.