Skip to content

Dataset · Roman1111111

Gemini 3.1 Pro synthetic reasoning chains

Roman1111111/gemini-3.1-pro-hard-high-reasoning

A compact 5.6M token corpus featuring complex multi-step English reasoning.

What's inside

The dataset consists of synthetic text generation and question-answering data generated by the Gemini 3.1 Pro model (High Reasoning variant). It contains between 1,000 and 10,000 examples (totaling 5.6M tokens) in English, collected with an emphasis on high logical density and multi-step verification of conclusions.

What you can do with it

  • Fine-tuning models or LoRA adapters to enhance reasoning skills (System 2)
  • Distilling logical reasoning chains into more compact LLMs
  • Testing models on complex tasks requiring step-by-step verification
Tasks
Question answering, Text generation
Languages
English
Size
1K–10K rows
Downloads this month
4.1K
Likes
63

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("Roman1111111/gemini-3.1-pro-hard-high-reasoning")