WebInstruct
220M instruction-response pairs extracted from the web by identifying educational content using a recall-then-verify pipeline. Covers K-12 through graduate-level content across math, science, and engineering — enabling large-scale instruction fine-tuning without human curation.
from datasets import load_dataset
ds = load_dataset("TIGER-Lab/WebInstructSub")Preview a sample row
{
"question": "What is the relationship between voltage, current, and resistance in Ohm's Law?",
"answer": "Ohm's Law states V = IR, where V is voltage (volts), I is current (amperes), and R is resistance (ohms). It means that for a given resistance, increasing voltage proportionally increases current..."
}Fine-tune with this dataset
Estimated VRAM to fine-tune with QLoRA (4-bit base model + LoRA adapters), using conservative defaults:
New to fine-tuning? Follow the step-by-step walkthrough: Fine-Tune Your First LLM in 1 Hour
Frequently asked questions
Can I use WebInstruct commercially?
Yes — WebInstruct is released under CC BY-SA 4.0, 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 WebInstruct contain, and do I need all of it?
WebInstruct contains 220M Pairs. 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 WebInstruct best used for?
Web-scale instruction mining for domain skills. It belongs to the Instruction / SFT section of our dataset hub, where you'll find alternatives and complementary sets.