Dataset · nvidia
Math problem-solving dataset with solutions and code
nvidia/OpenMathInstruct-1
1.8M problem-solution pairs with reasoning and code for math LLMs.
What's inside
OpenMathInstruct-1 contains 1.8M English problem-solution pairs. Problems are drawn from GSM8K and MATH train subsets, with solutions synthetically generated by Mixtral-8x7B featuring text reasoning and executable Python code blocks. Includes train and validation splits.
What you can do with it
- Training math LLMs
- LoRA fine-tuning for step-by-step solutions
- Evaluating math reasoning on GSM8K and MATH
- Tasks
- Question answering, Text generation
- Languages
- English
- Size
- 1M–10M rows
- Downloads this month
- 3.9K
- Likes
- 263
Models trained on it
- mradermacher/OpenMath-CodeLlama-7b-Python-hf-i1-GGUF 1.1K
- mradermacher/OpenMath-CodeLlama-13b-Python-hf-i1-GGUF 833
- Mungert/OpenMath-CodeLlama-7b-Python-hf-GGUF 678
- mradermacher/OpenMath-CodeLlama-34b-Python-hf-i1-GGUF 641
- Mungert/OpenMath-CodeLlama-13b-Python-hf-GGUF 608
- mradermacher/OpenMath-Llama-2-70b-hf-i1-GGUF 451
- mradermacher/OpenMath-CodeLlama-70b-Python-hf-i1-GGUF 416
- mradermacher/Llama-3.1-Swallow-8B-OpenMath-FT-i1-GGUFText generation 376
How to load it
from datasets import load_dataset
ds = load_dataset("nvidia/OpenMathInstruct-1")