Skip to content

Dataset · simplescaling

1,000 reasoning tasks from DeepSeek R1

simplescaling/s1K-1.1

A thousand questions with reasoning traces from DeepSeek R1 for training and validating reasoning models.

What's inside

A small English dataset of 1,000 questions, similar to s1K, but with solution traces generated by DeepSeek R1. The authors note that such traces yield noticeably better results. The sample structure includes a solution field with a step-by-step solution.

What you can do with it

  • Fine-tuning reasoning models on 1,000 examples
  • Validating DeepSeek R1 distillation
  • Evaluating models on short reasoning chains
Languages
English
Size
1K–10K rows
Downloads this month
39.4K
Likes
166

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("simplescaling/s1K-1.1")