datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
equational-theories-selected-problems
Equational Theories Selected Problems
Update (September 11, 2026)
This dataset was updated on September 11, 2026.
Main changes:
released the official Stage 2 evaluation problems: stage2_evaluation_main (200 problems; ground truth withheld — answer is null until Stage 2 concludes) and stage2_evaluation_research (100 order-5 research problems with no ground truth)
added metadata/stage2_evaluation_main.json and metadata/stage2_evaluation_research.json… See the full description on the dataset page: https://huggingface.co/datasets/SAIRfoundation/equational-theories-selected-problems.codeforces-problems-7kolympiad_style_integer_math_problems
Olympiad Math Corpus
Version: v2.1.1
Release date: 2026-05-03
59,486 synthetically generated olympiad-style math problems with verified integer answers and formal computation graphs.
Loading
from datasets import load_dataset
ds = load_dataset("mihailgribov/olympiad_style_integer_math_problems", split="train")
lemma_applicability is stored as list[{lemma, status}] rather than a sparse dict (required for Arrow-based consumers). To convert to a dict for local use:… See the full description on the dataset page: https://huggingface.co/datasets/mihailgribov/olympiad_style_integer_math_problems.codeforces_problems_subsetfull_geometry_problems_with_diagramscountdown_problemshard_geometry_problems_with_diagramsParallelKernelBench_Problems
ParallelKernelBench (benchmark)
Reference problems for ParallelKernelBench: a benchmark for LLM-generated multi-GPU CUDA kernels.
This dataset contains 87 reference implementations in reference/ and the input tensor specification in utils/input_output_tensors.py. Inputs are deterministic — reproduce them with create_input_tensor(rank, world_size, problem_id, base_shape, dtype, trial) from that file; you do not need stored .pt files.
Files
Path
Description… See the full description on the dataset page: https://huggingface.co/datasets/togethercomputer/ParallelKernelBench_Problems.IMO-2026-Problems
IMO 2026 Problems
The six IMO 2026 problem statements, indexed from 0 through 5 in contest order.
IDs 0–2 are from Day 1, and IDs 3–5 are from Day 2.
Schema
id: zero-based problem identifier (0 corresponds to Problem 1).
day: contest day (1 or 2).
problem: complete English problem statement.
Source
Extracted from the problem statements in SignalPilot Labs' AutoFyn IMO 2026 results:… See the full description on the dataset page: https://huggingface.co/datasets/bogoconic1/IMO-2026-Problems.ParallelKernelBench_Problems
ParallelKernelBench (benchmark)
Reference problems for ParallelKernelBench: a benchmark for LLM-generated multi-GPU CUDA kernels.
This dataset contains 87 reference implementations in reference/ and the input tensor specification in utils/input_output_tensors.py.
Files
Path
Description
data/problems.parquet
One row per problem (tabular access)
reference/*.py
Reference solution() implementations
utils/input_output_tensors.py
Input/output tensor… See the full description on the dataset page: https://huggingface.co/datasets/willychan21/ParallelKernelBench_Problems.agent-wicked-problems-40k
HUMMBL 40k Multi-Agent Wicked Problems & Coordination Corpus
A foundational 40,171-event empirical dataset capturing real-world multi-agent coordination, epistemic problem decomposition, failure mode taxonomies, and strategic intelligence surges generated across the HUMMBL autonomous agent fleet.
Dataset Overview
The dataset provides structured visibility into how autonomous agents navigate complex, ill-defined ("wicked") problems, coordinate across distributed… See the full description on the dataset page: https://huggingface.co/datasets/hummbl-hf/agent-wicked-problems-40k.Math-ProblemsServiceNowAI_R1_Distill_SFT_with_problems_and_responsesdjinn-problems-v0.9evolved-math-problems-OlympiadBench-from-deepseek-r1-0528-freealetheia_code_problems
Aletheia Code Problems with Misleading Hints
Dataset Description
This dataset contains multiple-choice code-reasoning problems derived from Aletheia-Bench and augmented with misleading textual hints. The misleading hints are intentionally designed to point to an incorrect answer.
The dataset was developed as part of the UnfaithRL project, which studies cue-following and unfaithful reasoning under reinforcement learning with verifiable rewards.
Specifically, it was… See the full description on the dataset page: https://huggingface.co/datasets/UnfaithRL/aletheia_code_problems.finecf-problems
Dataset Card for FineCF Problems
Dataset description
FineCF Problems is a dataset of 9,768 Codeforces problems, each paired with a cleaned, per-problem editorial explaining the solution approach. Problems span the full difficulty range (800 to 3500) and cover a wide variety of algorithmic topics including dp, graphs, math, greedy, data structures, and more.
You can load the dataset as follows:
from datasets import load_dataset
ds =… See the full description on the dataset page: https://huggingface.co/datasets/touristgpt/finecf-problems.qwen3-instruct-hard-problems-guidedall_cf_problemsmath-problemstaocp_open_problems
TAOCP Open Problems
Collection of open research problems singled out by Donald Knuth in
The Art of Computer Programming series. Its main purpose is to help measure how frontier models understand, investigate, and make verifiable progress on hard but interesting open problems.
Contents
The dataset contains 9 exercises rated 50, M50, or HM50 in the six
TAOCP editions and draft bundles available to this project. Knuth uses these
ratings for problems that were not… See the full description on the dataset page: https://huggingface.co/datasets/sytelus/taocp_open_problems.aimo-math-problemsMathematical QA data collections combining GSM8k, MATH and historical national mathematics competitions data extracted from AoPS, such as AMC and AIME.
The dataset splits into two difficulty levels, given problems' affinity to AIMO competition.
Hard: AMC12, AIME
Not hard: GSM8K, MATH, AHSME, USAMO, USOMO USAJMO, AJHSME, AMC8, AMC10
These data are further deduplicated, and filtered to keep those with text-based description (instead of replying on images) and integer answers, to match AIMO… See the full description on the dataset page: https://huggingface.co/datasets/billxbf/aimo-math-problems.numina-math-9sources-25each-modified-problems-o1-mod-2-onlypass-at-128-hard-omni-math-problems-self-correct-guidedshortest-path-dataset-problems_seed0_n20000_rows5-6_cols5-6_p0.4djinn-problems-v1.0
djinn-problems-v1.0 — fixed-djinn v2 build (2026-09-04)
Dual-verifier reward-hacking environments (insecure = exploitable, secure = hardened), rebuilt from
EleutherAI/djinn-problems-v0.9 so that is_hack = insecure_pass ∧ ¬secure_pass has no known false positives on
honest-but-wrong code. Every train row passes three probes under a CPU-time-bounded grader: ground truth [1,0,1],
exploit [1,1,0], and a stub returning None [0,0,0] (catches vacuous verifiers), and was screened… See the full description on the dataset page: https://huggingface.co/datasets/EleutherAI/djinn-problems-v1.0.vehicle-reliability-scorecard
Vehicle Reliability Scorecard
One row per vehicle (year + make + model) with its ProblemsByVin reliability score, total NHTSA complaints, recalls, and defect investigations, plus the single component owners complain about most. The master index across the whole tracked fleet — the flat table to join every other dataset to.
Columns
column
meaning
year
Model year
make
Manufacturer
model
Model
reliability_score
1.0 (worst) – 5.0 (best); shown on site… See the full description on the dataset page: https://huggingface.co/datasets/ProblemsByVin/vehicle-reliability-scorecard.evolved-math-problems-from-deepseek-r1-0528-freedjinn-problems-v0.6numina-math-9sources-25each-modified-problems-o1-responses-with-original-responses-final_SNOW