An environement builder for hierarchical reasoning research

PyTorch implementation of the Hierarchical Reasoning Model with Adaptive Computation Time (HRMACT) for solving Sudoku puzzles, with pretrained checkpoints available on Hugging Face . Supports training, inference, and evaluation with curriculum learning across difficulty levels.

A theoretical framework for constructing adaptive, hierarchical layers of knowledge processing in AI. This model enables dynamic restructuring of knowledge systems, supporting scalable learning and meta-reasoning. AIにおける知識処理を階層的かつ適応的に構築する理論モデル。知識体系の動的再構成を可能にし、スケーラブルな学習とメタ認知的推論を支援する。