Skip to content

Dataset · glaiveai

22M synthetic reasoning traces

glaiveai/reasoning-v1-20m

Synthetic question-answer pairs with reasoning chains on general topics, beyond just math and code.

What's inside

The dataset contains over 22 million questions and reasoning-backed answers generated by the deepseek-ai/DeepSeek-R1-Distill-Llama-70B model. Topics span social and natural sciences, education, creative writing, and general conversation; the language is English. The format consists of text question-answer pairs for generative training.

What you can do with it

  • SFT models on reasoning traces outside of code and math
  • Evaluating LLM generalization on humanities topics
  • LoRA fine-tuning for conversational reasoning
Tasks
Text generation
Languages
English
Size
10M–100M rows
Downloads this month
6.9K
Likes
245

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("glaiveai/reasoning-v1-20m")