Dataset · hoskinson-center
Proof-Pile math text corpus
hoskinson-center/proof-pile
High-quality English math texts for language model training and fine-tuning.
What's inside
Dataset contains English math texts; metadata size ranges from 100k to 1M examples. The card only indicates high-quality math text; sources and exact format are not described.
What you can do with it
- Fine-tuning LLMs to solve math problems
- SFT of models on mathematical reasoning
- Evaluating the quality of mathematical text generation
- Tasks
- Text generation
- Languages
- English
- Size
- 100K–1M rows
- Downloads this month
- 4.5K
- Likes
- 70
Models trained on it
- BSC-LT/salamandra-7b-instructText generation 31.5K
- BSC-LT/salamandra-7bText generation 3.7K
- BSC-LT/salamandra-2bText generation 3.6K
- BSC-LT/salamandra-2b-instructText generation 2.2K
- BSC-LT/ALIA-40bText generation 2K
- mradermacher/ALIA-40b-i1-GGUF 1.1K
- mradermacher/ALIA-40b-GGUF 492
- mradermacher/salamandra-2b-instruct-GGUF 426
How to load it
from datasets import load_dataset
ds = load_dataset("hoskinson-center/proof-pile")