No Robots — LLM 指令 / SFT Dataset
10,000 instructions and demonstrations written entirely by skilled human annotators — zero synthetic data. Modeled after the SFT data described in OpenAI's InstructGPT paper, split into 9.5k train / 500 test examples across categories like generation, open QA, brainstorming, and coding. Small, clean, and diverse: widely considered the best first dataset for learning to fine-tune.
Dataset Details
| Provider | HuggingFace H4 |
| Category | 指令 / SFT |
| Size | 10k Rows |
| License | CC BY-NC 4.0 |
| Downloads | n/a |
| Tags | Human-Written, No-Synthetic, SFT, Beginner-Friendly |
from datasets import load_dataset
ds = load_dataset("HuggingFaceH4/no_robots")
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
- UltraFeedback — The default DPO preference set to run after any SFT pass
- Databricks Dolly 15k — Commercially safe instruction tuning - fully human-written
- Python-Edu — Continued pretraining for Python code understanding
- OpenHermes 2.5 — The default general-purpose SFT mix for 7B-13B fine-tunes
常见问题
No Robots 可以商用吗?
不能用于产品——No Robots 采用 CC BY-NC 4.0,仅限研究等非商业用途。商业微调请改用同类别中宽松许可的数据集。
No Robots 有多少数据?需要全部使用吗?
No Robots 包含 10k 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。
No Robots 最适合做什么?
Your first fine-tune — small, clean, 100% human-written SFT data (non-commercial license)。它属于数据集中心的「指令 / SFT」板块,那里有替代和互补的数据集。
← 全部数据集 | Fine-Tuning Guide