A Machine Learning Approach That Beats Large Rubik's Cubes. Weights and Datasets

This dataset contains model weights and evaluation data used in experimental studies of the "A Machine Learning Approach That Beats Large Rubik's Cubes" research. solvers/ Each subdirectory corresponds to a solver model from Table 1 of the paper. These models were trained for different cube sizes (2x2x2, 3x3x3, 4x4x4, 5x5x5). Each solver folder contains: - weights/ – Model weights in .pth format (Float32), compatible with model.py. The filenames follow this pattern: <model architecture number from Table 2>-<agent ID>-<train set size (optional)>.pth - If multiple agents were trained with the same model architecture, they have different ID numbers. - The train set size is included only if different versions exist. - dataset/ - generators.json – Defines permutation actions and symbolic names used in searcher.py. - data.pt – A 2D Torch tensor (Int8) where each row represents a scrambled Rubik’s Cube state. Values in [0, 6) correspond to face colors. figures/ Each subdirectory (fig-*) corresponds to an experiment from the paper. The structure inside each folder (weights/ and dataset/) is the same as in solvers/, containing model weights and evaluation data.