A high-quality 2025 synthetic dataset created to train the SmolLM2 family of small language models. 1 million examples covering diverse tasks, generated with careful quality filtering. Demonstrates that small models (1.7B, 360M) can achieve strong performance with the right training data.
Dataset Details
Provider
HuggingFaceTB
Category
Instruction / SFT
Size
1M Rows
License
Apache 2.0
Downloads
600k
Tags
Synthetic, 2025, Small-Models, High-Quality
from datasets import load_dataset
ds = load_dataset("HuggingFaceTB/smoltalk")
Fine-tune with this dataset
Estimated VRAM to fine-tune with QLoRA (4-bit base model + LoRA adapters), using conservative defaults:
SmolTalk 2 — Reproducing a complete modern post-training pipeline (mid-training → SFT → preference) for
No Robots — Your first fine-tune — small, clean, 100% human-written SFT data (non-commercial license)
Python-Edu — Continued pretraining for Python code understanding
OpenHermes 2.5 — The default general-purpose SFT mix for 7B-13B fine-tunes
Frequently asked questions
Can I use Smoltalk commercially?
Yes — Smoltalk is released under Apache 2.0, a permissive license that allows commercial use, including training models you ship in a product. Check the dataset card for attribution requirements before release.
How much data does Smoltalk contain, and do I need all of it?
Smoltalk contains 1M Rows. You rarely need all of it: for style and format fine-tuning, a few hundred to a few thousand examples are enough — load a slice (e.g. split="train[:1000]") and scale up only if quality plateaus.
What is Smoltalk best used for?
General SFT for small models (the SmolLM2 recipe). It belongs to the Instruction / SFT section of our dataset hub, where you'll find alternatives and complementary sets.