High-Reasoning Dataset v1
Abstract
2,139 multi-AI knowledge-distilled Q&A pairs with deep chain-of-thought, math proofs, production code and edge cases. Built by an autonomous two-model loop: one asking relentless questions, one teaching. MIT licensed.
Read on Hugging Face 
On this page
Abstract
2,139 multi-AI knowledge-distilled Q&A pairs with deep chain-of-thought, math proofs, production code and edge cases. Built by an autonomous two-model loop: one asking relentless questions, one teaching. MIT licensed.
Overview
High-Reasoning Dataset v1 is a small, dense distillation corpus. Two models ran continuously: one generating hard questions, one answering with full reasoning. The output was filtered for depth, correctness and non-triviality.
The result is 2,139 pairs covering multi-step math, proofs, production-grade code and edge cases that usually get skipped.
Composition
| Property | Value |
|---|---|
| Rows | 2,139 |
| Format | Q&A pairs with chain-of-thought |
| Domains | Math, proofs, production code, edge cases |
| Construction | Multi-AI knowledge distillation |
| License | MIT |
How it is meant to be used
- As a high-quality SFT supplement, not a bulk pretraining corpus.
- As an evaluation set for reasoning depth on a small number of hard prompts.
- As a seed corpus for further distillation with a stronger teacher.
- Not as ground truth: distilled answers can still be wrong and should be verified before training on them at scale.
Cite this
@dataset{wakharkar_high_reasoning_v1,
title = {High-Reasoning Dataset v1},
author = {Aditya Wakharkar},
year = {2026},
license = {MIT},
url = {https://huggingface.co/datasets/tantra-ai-labs/high-reasoning-dataset-v1}
}The canonical version of this work lives on Hugging Face. If a number here and the card disagree, the card wins.