OpenThoughts3-1.2M — LLM 推理 Dataset
1.2M reasoning traces — 850k math, 250k code, and 100k science questions — with chains of thought generated by QwQ-32B. The result of 1,000+ controlled experiments on reasoning-data curation, and the training set behind OpenThinker3-7B, the state-of-the-art open-data 7B reasoning model (53% AIME 2025). The reference open recipe for distilling reasoning ability.
Dataset Details
| Provider | Open Thoughts |
| Category | 推理 |
| Size | 1.2M Rows |
| License | Apache 2.0 |
| Downloads | n/a |
| Tags | Reasoning-Traces, Math, Code, Science, Distillation, 2025 |
from datasets import load_dataset
ds = load_dataset("open-thoughts/OpenThoughts3-1.2M")
用这个数据集微调
使用 QLoRA（4-bit 基础模型 + LoRA 适配器）微调的预计显存需求（保守默认参数）：
| 7B QLoRA | ~6GB VRAM |
| 13B QLoRA | ~10GB VRAM |
检测你的 GPU 能否微调 →
微调新手？跟着分步教程走： 一小时微调你的第一个 LLM
相关数据集
- NuminaMath — Chain-of-thought math fine-tuning up to olympiad level
- OpenCodeReasoning — Fine-tuning coding models that show their reasoning before writing the solution
- MetaMathQA — Boosting GSM8K/MATH-style math skills in 7B models
- Orca Math Word Problems — Grade-school math word problems for small models
常见问题
OpenThoughts3-1.2M 可以商用吗？
可以——OpenThoughts3-1.2M 采用 Apache 2.0 宽松许可证，允许商业使用，包括训练用于产品的模型。发布前请查看数据集卡片中的署名要求。
OpenThoughts3-1.2M 有多少数据？需要全部使用吗？
OpenThoughts3-1.2M 包含 1.2M Rows。通常不需要全部：风格和格式微调只需几百到几千条样本——先加载切片（如 split="train[:1000]"），质量到达瓶颈时再扩大规模。
OpenThoughts3-1.2M 最适合做什么？
Distilling strong math/code/science reasoning into 7B–32B models。它属于数据集中心的「推理」板块，那里有替代和互补的数据集。
← 全部数据集 | Fine-Tuning Guide