Dataset · MiniMaxAI
Synthetic logic tasks with automatic verification
MiniMaxAI/SynLogic
35 types of logic tasks with automated answer checking for RL with verifiable rewards.
What's inside
SynLogic is a synthetic dataset for developing LLM logical reasoning through reinforcement learning with verifiable rewards. It contains 35 different logic tasks with automated checking. Languages are English and Chinese, size is from 10K to 100K examples.
What you can do with it
- RLVR training of LLMs for logical reasoning
- Evaluation and comparison of reasoning models across 35 task types
- SFT/LoRA for a model solving logic tasks with a verifiable answer
- Tasks
- Text generation
- Languages
- English, Chinese
- Size
- 10K–100K rows
- Downloads this month
- 1.5K
- Likes
- 109
Models trained on it
- MiniMaxAI/SynLogic-7BText generation 2.4K
- mradermacher/SynLogic-7B-GGUF 968
- MiniMaxAI/SynLogic-32BText generation 966
- mradermacher/Megatron-Bots-1.7B-Reasoning-GGUF 910
- Jonjew/NSFWMasterText → image 700
- mradermacher/SynLogic-32B-i1-GGUF 600
- mradermacher/SynLogic-Mix-3-32B-i1-GGUF 583
- mradermacher/SynLogic-Mix-3-32B-GGUF 564
How to load it
from datasets import load_dataset
ds = load_dataset("MiniMaxAI/SynLogic")