AtlasUnified/atlas-math-sets-2.0

Atlas Math Sets 2.0

Atlas Math Sets 2.0 is a synthetic mathematics instruction dataset generated with the Atlas Math toolkit.

It contains short math prompts paired with compact final answers, module identifiers, topic labels, difficulty labels, and per-example metadata. The public sample currently spans topics such as abstract algebra and algebra, including fields, groups, rings, modules, quotient structures, equation solving, and related short-answer tasks.

What this dataset is

• Synthetic math instruction data
• Short-form question answering and answer generation
• Topic- and subtopic-labeled examples
• Generator-defined difficulty levels
• Metadata-rich records for filtering and analysis

What this dataset is not

• A human-authored tutoring corpus
• A proof-level reasoning benchmark
• A replacement for broad mathematical evaluation
• A guarantee of natural student phrasing or real classroom distributions

Dataset Structure

Each row is a JSON-style record. The visible public schema is:

{
  "module_id": "abstract_algebra.fields_modules.field_extensions_intro",
  "topic": "abstract_algebra",
  "subtopic": "fields_modules.field_extensions_intro",
  "difficulty": "level_1",
  "instruction": "Compute If E contains F as a subfield, what is E over F called?.",
  "input_text": "If E contains F as a subfield, what is E over F called?",
  "answer": "a field extension",
  "metadata": {"concept": "definition"}
}

Fields

Field	Description
module_id	Full generator/module identifier.
topic	Top-level math area, such as abstract_algebra or algebra.
subtopic	More specific module category.
difficulty	Generator-defined level, such as level_1, level_2, or level_3.
instruction	Instruction-style prompt shown to the model.
input_text	Core math question, expression, or problem text.
answer	Canonical short answer string.
metadata	Module-specific structured details used for filtering or analysis.

Example Records

{"module_id":"abstract_algebra.fields_modules.field_extensions_intro","topic":"abstract_algebra","subtopic":"fields_modules.field_extensions_intro","difficulty":"level_1","instruction":"Compute If E contains F as a subfield, what is E over F called?.","input_text":"If E contains F as a subfield, what is E over F called?","answer":"a field extension","metadata":{"concept":"definition"}}
{"module_id":"abstract_algebra.groups.binary_operations","topic":"abstract_algebra","subtopic":"groups.binary_operations","difficulty":"level_3","instruction":"Classify the operation in On positive integers, define a*b=a-b. Is this a binary operation?.","input_text":"On positive integers, define a*b=a-b. Is this a binary operation?","answer":"no","metadata":{"set":"positive integers","closed":false}}
{"module_id":"algebra.equations.multi_step","topic":"algebra","subtopic":"equations.multi_step","difficulty":"level_1","instruction":"Solve the multi-step equation 4x + -3 = -14 - 5.","input_text":"4x + -3 = -14 - 5","answer":"-4","metadata":{"step_count":3,"has_variable_both_sides":false}}

Splits

The dataset exposes three splits:

• train
• validation
• test

Intended Uses

• Supervised fine-tuning on compact math prompts
• Short-answer math evaluation
• Topic/subtopic filtering experiments
• Difficulty-conditioned curriculum training
• Synthetic data generation and deduplication experiments

Limitations

• The data is synthetic and generator-shaped.
• Difficulty labels come from generation logic, not necessarily human calibration.
• Many examples expect concise final answers rather than full derivations.
• Strong performance here may not transfer to open-ended math reasoning.
• Metadata schemas can vary by module.

Loading

from datasets import load_dataset

ds = load_dataset("AtlasUnified/atlas-math-sets-2.0")
print(ds)
print(ds["train"][0])

Basic Prompt Format

row = ds["train"][0]

prompt = f"Instruction: {row['instruction']}\nInput: {row['input_text']}\nAnswer:"

Source

Generated with the Atlas Math toolkit:

• GitHub: atlasunified/atlas-math
• Hugging Face dataset: AtlasUnified/atlas-math-sets-2.0

Citation

@dataset{atlas_math_sets_2,
  title  = {Atlas Math Sets 2.0},
  author = {AtlasUnified},
  year   = {2026},
  note   = {Synthetic mathematics instruction dataset generated with Atlas Math}
}

License

MIT