Datasets:

---

SalutefaceAAC

/

math-word-problems-graded

Like 0

math-word-problems-graded v1.0 is live!

Hi HuggingFace community! 👋

I just published math-word-problems-graded, a synthetic dataset of 1,300 graded math word problems with step-by-step solutions, perfect for training and evaluating mathematical reasoning in language models.

What's in the dataset?

Topics covered: Arithmetic, Algebra, Geometry, Probability, Statistics, Calculus

Difficulty levels: easy → medium → hard → expert

Grade levels: Elementary → Middle School → High School → College

Schema

Each record contains:

• problem — the word problem in natural language
• solution — step-by-step solution showing all work
• answer — the final answer
• difficulty, grade_level, topic, domain — rich metadata for filtering

Use cases

• Fine-tuning for GSM8K-style mathematical reasoning
• Curriculum learning experiments (easy → hard)
• Multi-step reasoning benchmarking
• Educational QA / tutoring systems

Quick start

from datasets import load_dataset
ds = load_dataset("SalutefaceAAC/math-word-problems-graded")

# Filter by difficulty
hard_problems = ds["train"].filter(lambda x: x["difficulty"] == "hard")

# Filter by topic
geometry = ds["train"].filter(lambda x: x["topic"] == "geometry")

Dataset card with full documentation is in the README. Feedback, PRs, and issues welcome!

Built as part of the SalutefaceAAC OSS dataset initiative — contributing synthetic datasets to underserved corners of the HuggingFace ecosystem.