metadata

title: Rubik Tensor
emoji: ⚡
colorFrom: blue
colorTo: blue
sdk: docker
pinned: true
license: apache-2.0
short_description: Interface for playing with Rubik cubes of any size

Rubik-Tensor

See the HF Space JBAujogue/Rubik-Tensor (make sure to accept 3rd-party cookies through the Eye icon of the URL bar).

Setup

This project uses uv 0.8.3 as environment & dependency manager, and python 3.11 as core interpreter. Install the project with

uv venv
(Activate env)
uv sync --extra [extra]
pre-commit install

where extra should be one of the following:

torch: pytorch package released on pypi (cpu-only form non-linux systems, cuda-enabled for linux systems).
torch-cpu: cpu-only torch wheel.
torch-cu126: cuda 12.6-compatible torch wheel.

Usage

Launch the web interface

python -m rubik interface

Use the python API

from rubik.cube import Cube

cube = Cube(size=3)

# display cube state
print(cube)
#     UUU        
#     UUU
#     UUU
# LLL CCC RRR BBB
# LLL CCC RRR BBB
# LLL CCC RRR BBB
#     DDD
#     DDD
#     DDD

# display history of moves
print(cube.history)
# []

# scramble the cube using 1000 random moves (this resets the history)
cube.scramble(num_moves=1000, seed=0)

# rotate it in some way (this gets appended to history)
cube.rotate('X2 X1i Y1i Z1i Y0 Z0i X2 X1i Y1i Z1i Y0 Z0i')

Roadmap

Fully tensorized Rubik Cube model

☑️ Tensorized states.
☑️ Tensorized actions.
☑️ Interface for performing actions.

Movement explorer

⬜ Explore changes resulting from a sequences of moves.
⬜ Find least sequences of moves satisfying some input constrains.

Visualization interface

Base solvers following rule-based policies

References

Implementations & rule-based solvers

Open-source projects related to Rubik's Cube:

Machine Learning based solver models

2025, CayleyPy Cube, Github, Paper
2025, Solving A Rubik’s Cube with Supervised Learning – Intuitively and Exhaustively Explained, Blog post
2024, Solving Rubik's Cube Without Tricky Sampling, Paper.
This involves training a scorer, that estimates the number of moves transforming a given source state into a given target state, where the latter is not necessarily a solved cube. Data are synthetically generated performing random moves and factorizing repeated identical moves.
2023, Curious Transformer, Github
2021, Efficient Cube, Github, Paper