Dataset · Qwen
Qwen3Guard model reasoning safety benchmark
Qwen/Qwen3GuardTest
Dataset for evaluating the safety of both final answers and intermediate reasoning steps of models.
What's inside
A test benchmark from the Qwen team for evaluating Qwen3Guard moderation systems. It contains 1,000 to 10,000 examples in text and tabular formats. Unlike standard safety tests, this evaluates models' long reasoning chains.
What you can do with it
- Evaluating and testing safety guardrails for LLMs
- Checking safety of intermediate steps in reasoning models
- Comparing content filters on complex scenarios
- Tasks
- Text classification
- Size
- 1K–10K rows
- Downloads this month
- 792
- Likes
- 37
How to load it
from datasets import load_dataset
ds = load_dataset("Qwen/Qwen3GuardTest")