A dataset for training and evaluating logical reasoning capabilities in LLMs through constraint satisfaction problems.
Domain: Multi-step logical reasoning / constraint satisfaction / schedulingSize: 1000 examples (uniquely solvable puzzles with step-by-step reasoning)Format: Natural language constraint satisfaction puzzles with structured solutionsGeneration: Algorithmically created with guaranteed… See the full description on the dataset page:
https://huggingface.co/datasets/strickvl/constraint-sat-1000.