The Arithmetic Word Problem Compendium
A high-quality, synthetic dataset for training AI/LLMs for deep reasoning. 10 million unique, mathematical word problems.
Highlights
High-Quality Data
Precise, accurate math across 7 domains.
Grammatical correctness and rich vocabulary, with a templating system carefully validated by human authors.
Step-by-step solutions accompany each problem.
Tagged with metadata.
Time Savings
Save researchers and developers time and effort compared to creating their own.
Comes with both training and evaluation sets.
Clean, original data, never before seen online.
Use as a benchmark to evaluate your models.
Improved AI Performance
Our data can enhance the reasoning and problem-solving capabilities of LLMs.
Contains deep process reasoning through multiple steps of logic.
Create high-performance LLMs with enhanced reasoning.
Use in pertaining, instruction tuning, fine-tuning, and distilling
Tiers of data offered.
Start with our free sample - inspect the contents of our dataset across the all different domains.
Then contact us about:
Getting access to 100,000 word problems, all unique.
Getting the full 10 million word problems.
Licensing our human-validated, scalable templating system for magnitudes more data, and to customize the number of operations in each problem to as many as you want for deep reasoning.
About
our company
We have more than a decade in experience with software development, machine learning experience, and professional writing. Based in Austin, TX.
