s1K-1.1 — LLM 推理 Dataset

Just 1,000 ultra-curated questions with reasoning traces — selected from 59k candidates for difficulty, diversity, and quality, with traces regenerated by DeepSeek R1 in the 1.1 release. Fine-tuning Qwen2.5-32B on s1K (26 minutes on 16 H100s) plus 'budget forcing' at inference exceeded o1-preview on competition math. The LIMA of reasoning: proof that data quality can beat quantity.

Dataset Details

from datasets import load_dataset
ds = load_dataset("simplescaling/s1K-1.1")

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）：

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

• LIMA: Less Is More for Alignment — Style and format alignment with a tiny curated set - quality over quantity
• OpenThoughts3-1.2M — Distilling strong math/code/science reasoning into 7B–32B models
• 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

← 全部数据集 | Fine-Tuning Guide