Dataset · TIGER-Lab
MMLU-Pro: challenging knowledge benchmark
TIGER-Lab/MMLU-Pro
12k complex multi-discipline questions for rigorous LLM evaluation.
What's inside
The dataset contains 12k complex questions across various disciplines. It is a more robust and difficult set for multi-task understanding and LLM benchmarking. Modalities are tabular and text, English language; size ranges from 10k to 100k examples.
What you can do with it
- Evaluating LLMs on MMLU-Pro
- Comparing models on the leaderboard
- Testing reasoning after SFT
- Tasks
- Question answering
- Languages
- English
- Size
- 10K–100K rows
- Downloads this month
- 397.1K
- Likes
- 607
Models trained on it
- vllm-sr/mmbert32k-intent-classifier-mergedText classification 3K
- mradermacher/Atomight-V2.2-0.5B-UltraThink-GGUF 559
- raghul2004/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-Ternary-2bitText generation 442
- Merlinoz11/extGemma4-44B-GGUF 427
- mradermacher/Atomight-V2.2-UltraThink-0.5B-GGUF 351
- notfakerv/round2-1.7b-loop-previewText generation 218
- usermma/Atomight-V2.2-0.5B-UltraThink-GGUF-mradermacher 198
- OpceanAI/Qwen3.5-0.8B-abliteratedImage text → text 141
How to load it
from datasets import load_dataset
ds = load_dataset("TIGER-Lab/MMLU-Pro")