Dataset · truthfulqa
TruthfulQA: LLM truthfulness benchmark
truthfulqa/truthful_qa
817 questions across 38 categories to test whether a model generates truthful answers.
What's inside
TruthfulQA is a benchmark that measures how truthfully a language model answers questions. It includes 817 questions spanning 38 categories, including health, law, finance, and politics. The questions are crafted so that humans might answer falsely due to misconceptions, testing whether models avoid repeating learned falsehoods.
What you can do with it
- Evaluating language model truthfulness
- Testing QA models
- Comparing models on the benchmark
- Tasks
- Multiple choice, Text generation, Question answering
- Languages
- English
- Size
- 1K–10K rows
- Downloads this month
- 96.7K
- Likes
- 302
Models trained on it
- mradermacher/bobcat-flash-1.2-GGUF 1.1K
- mradermacher/Safe4Qwen-1.7B-i1-GGUF 969
- katanemo/Arch-GuardText classification 396
- pockettravel/qwen3.5-2b-travel-en-GGUFText generation 328
- pockettravel/qwen3.5-0.8b-travel-en-GGUFText generation 295
- pockettravel/qwen3.5-4b-travel-en-GGUFText generation 240
- mradermacher/Safe4Qwen-1.7B-GGUF 215
- katanemo/Arch-Guard-cpuText classification 105
How to load it
from datasets import load_dataset
ds = load_dataset("truthfulqa/truthful_qa")