Skip to content

Dataset · google

IFEval instruction-following benchmark

google/IFEval

Around 500 verifiable instructions for evaluating how well LLMs follow explicit constraints.

What's inside

A prompt dataset from the Instruction-Following Eval (IFEval) benchmark for large language models. Contains around 500 verifiable instructions, such as writing over 400 words or mentioning AI at least three times. Execution verification relies on heuristics. English language.

What you can do with it

  • Evaluate LLMs on instruction following
  • Compare base and fine-tuned models
  • Test LoRA after SFT
Tasks
Text generation
Languages
English
Size
up to 1K rows
Downloads this month
190.6K
Likes
190

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("google/IFEval")