A "CIFAR-for-Reasoning" dataset designed to test sequence models on Deduction, Induction, and Abduction tasks with strict Chain-of-Thought supervision.
This dataset implements the paradigm proposed in "Absolute Zero: Reinforced Self-play Reasoning with Zero Data" (Zhao et al., 2025), but adapted for the deterministic environment of Elementary Cellular Automata (Rule 0-255). It isolates algorithmic reasoning capabilities… See the full description on the dataset page:
https://huggingface.co/datasets/kreasof-ai/ECA-Zero.
