Datasetpublic

Emulated-Inc/olympiad-math-training-pool

Olympiad mathematics training pool Public olympiad and competition mathematics, four datasets gathered at pinned revisions, shipped twice over. sources/ holds each dataset the way its publisher ships it, in its own file format with its own fields and nothing renamed, 229052 rows across four folders. pool/ holds the union of those same datasets in one format, one JSON object per line, deduplicated by problem text and reduced to 225822 rows, every row labelled with the dataset it… See the full description on the dataset page: https://huggingface.co/datasets/Emulated-Inc/olympiad-math-training-pool.

sourceHugging Faceapache-2.0updated 11d agoView on Hugging Face

0likes668downloads

Dataset Card

Olympiad mathematics training pool

Public olympiad and competition mathematics, four datasets gathered at pinned revisions, shipped twice over. sources/ holds each dataset the way its publisher ships it, in its own file format with its own fields and nothing renamed, 229052 rows across four folders. pool/ holds the union of those same datasets in one format, one JSON object per line, deduplicated by problem text and reduced to 225822 rows, every row labelled with the dataset it came from, that dataset's revision and its provenance class.

Nothing here is filtered for quality, for difficulty or for how it was produced. Problems written by people, problems copied from competition papers, and solutions a language model rewrote all sit in the pool, labelled, and which of them to use is the reader's decision. Take the union if one format is what you want, take a source folder if you already know that dataset and want its own columns.

Layout

README.md
pool/pool-00000-of-00003.jsonl  ...  pool-00002-of-00003.jsonl   the union, 225822 rows
sources/deepscaler/              1 file,   36695 rows
sources/olympiadbench_other/    11 files,   5000 rows
sources/omni_math/               1 file,    2977 rows
sources/numinamath_olympiads/    3 files, 184380 rows

The union is shuffled with a fixed seed, so any prefix of any file is a sample of the whole pool rather than of one source. The pool is about 548 MB on disk, 365 MB of it the union and 184 MB the source folders.

The union under pool/

One JSON object per line, with these fields.

Field	What it holds
`id`	`olympiad_` and a seven digit number, unique across the whole union
`problem`	the problem text, exactly as the source publishes it, LaTeX and all
`answer`	the final answer alone, as a string, or null when the row has none
`solution`	the worked solution, or null when the row has none
`has_answer`	whether `answer` is not null
`has_solution`	whether `solution` is not null
`answer_from`	`dataset` when the source published the answer as its own field, `boxed_in_solution` when it was read from the last box of the solution, null when there is no answer
`source`	the folder under `sources/` the row came from
`source_subset`	the value of that source's own subset column, where it has one, and the configuration name for the one source that splits by file
`source_repo`	the Hugging Face repository the source was taken from
`source_revision`	the commit of that repository, so every row is traceable to a pinned revision
`source_file`	the file under `sources/` the row came from
`source_row`	the row's position in that file, counting from zero
`provenance`	how the PROBLEM was produced: `human`, `collected` or `model-generated`
`solution_provenance`	how the SOLUTION was produced, on the same three values, null when there is no solution
`language`	`en` or `zh`

Counts over the 225822 rows: 151316 carry an answer and 195341 carry a solution. Of the answers, 151307 are the source's own answer field and 9 were read out of the last box of a solution. By provenance of the problem: 217623 collected, 8199 human, none model-generated. By language: 221281 English, 4541 Chinese.

By source: 184335 from numinamatholympiads, 35809 from deepscaler, 4998 from olympiadbenchother and 680 from omnimath. That last number is small because the union keeps one row per distinct problem and prefers the copy that carries the most, an answer first, then a solution, and then the earlier source in the list above. Most omnimath problems are also in deepscaler, which was built in part from them, so they lose that comparison, and the full omni_math file sits under sources/.

The largest subsets are olympiads with 181814 rows and olympiads_ref with 2521, both from numinamatholympiads, then deepscaler with 35809, then the Chinese entrance-examination configurations of olympiadbenchother with 3839 between them.

The sources under sources/

deepscaler, agentica-org/DeepScaleR-Preview-Dataset at b6ae8c60f5c1f2b594e2140b91c49c9ad0949e29, 36695 rows in one JSON file holding a list of objects with problem, answer and solution, the answer being the final answer alone. MIT. Provenance collected, solutions collected. About forty thousand competition problems with a single short answer, compiled from the AIME papers of 1984 to 2023, the AMC papers before 2023, an olympiad collection and the STILL collection. The problems and the worked solutions are the ones the competitions and the archives published. A model was used only to read the final answer out of each solution, and answers a symbolic checker could not handle were dropped.

olympiadbench_other, Hothan/OlympiadBench at 91184b52131e7fc9455fef848035173aea8cc01a, 5000 rows across eleven parquet files, one per configuration, with question, solution (a list), finalanswer (a list), context, unit, answertype, subfield, language and the image columns, which carry data only in the MM configurations. Apache 2.0. Provenance human, solutions human. Olympiad and Chinese college entrance examination mathematics, with expert step-by-step solutions written for the collection. One file per configuration: the name says open-ended (OE) or theorem proof (TP), text-only (TO) or with figures (MM), the language and whether the problems come from a competition (COMP) or the entrance examination (CEE). The theorem-proof files carry a solution and no final answer, and the figure files carry the figures themselves in their image columns, which means their problems refer to pictures this pool does not reproduce. One configuration of the twelve that repository publishes for mathematics is not here.

omni_math, KbsdJames/Omni-MATH at 40ba231d8f16e29ecd40e6407e2c8640145a8f62, 2977 rows, one JSON object per line with problem, solution, answer, domain, difficulty and source, the source naming the competition and the difficulty being a rating from 1 to 10. Apache 2.0. Provenance human, solutions human. Olympiad problems collected from the official competition websites and the Art of Problem Solving wiki, with the solutions those sites published and a difficulty rating and a subject path added by hand. Problems and solutions are both written by people. AMC and AIME problems were deliberately left out when it was built.

numinamath_olympiads, AI-MO/NuminaMath-1.5 at 1b05109f9e5c1ad06c0663519502416c30b300f8, 184380 rows across three parquet shards with problem, solution, answer, problemtype, questiontype, problemisvalid, solutionisvalid, source and synthetic. An answer of proof or notfound means the row has no short answer. Apache 2.0. Provenance collected, solutions model-generated. This is the olympiad part of the largest compilation of competition mathematics on the hub: the rows whose source column says olympiads, drawn from olympiad papers, and the rows whose source column says olympiads_ref, which the compilers parsed and checked by hand against the official websites of national olympiads. The other subsets of that compilation, its school material and the problems a model wrote from seed problems, are not here, which is why these shards are far smaller than the ones that repository publishes. The solutions were rewritten into a uniform style by a large language model whatever the problem's origin, so the solution text is model-generated across the file, and the short answers were extracted by the same model.

Provenance in plain words

—human: a person wrote the text, with no model in the loop.
—collected: the text is a real artefact of people doing something, a competition paper, an examination, a forum post, a textbook, gathered rather than written for a dataset.
—model-generated: a language model wrote it. No problem in this pool is in this class. Every solution in numinamath_olympiads is, because that compilation rewrote all of its solutions with a model.
—rule-generated: a program or a template produced it. Nothing in this pool is in this class.

Most of the pool is collected: real olympiad and competition problems, gathered at scale. The human class is the two collections whose problems and solutions people wrote and checked. If you want only problems people wrote or sat, every row here qualifies. If you also care that the reasoning was written by a person rather than rewritten by a model, filter solution_provenance to human and collected, which leaves the omnimath, olympiadbenchother and deepscaler rows.

Licences

Each source is redistributed here under the licence its publisher declares for it, and the licence of every row is the licence of its source folder.

Source	Licence
deepscaler	MIT
olympiadbench_other	Apache 2.0
omni_math	Apache 2.0
numinamath_olympiads	Apache 2.0

Apache 2.0 is the more demanding of the two, so it governs the pool as a whole. No source here is NonCommercial and none is untagged. The compilations gather material from many places, and the licence above is the one their publisher applies to the compilation.

Three corpora that would otherwise belong here are deliberately absent. The MATH training split is not included: the original upload was disabled on a takedown notice from the rightsholder, which also asked for datasets branched off it to be removed, and of the mirrors that remain one declares no licence at all, so no clean copy exists to redistribute. The Art of Problem Solving forum instruction corpus is not included either: its authors disclaim the licence tag the upload carries, and the forum it was crawled from now publishes a machine-readable reservation of rights against training use. The reasoning-trace corpus built on the compilation above is not included because its problems are already here and everything it adds is model-written reasoning over them.

A note on the counts

The counts above are the counts in this pool. Rows were removed from every source while it was assembled, so they do not match the totals the source repositories publish.