Skip to content

Conversation

mohabedalgani
Copy link

What

  • One-click Colab notebook that reproduces the Sudoku-Extreme 1 k experiment on a free T4 GPU (~55 min).
  • Fixes #0 (no open issue, but fills a gap in quick-start docs).

How tested

  • Ran end-to-end in Colab PRO high-ram today → final exact accuracy: 99.3 % (see notebook).
  • Badge uses relative link so it will work after merge.

Files changed

  • notebooks/colab/HRM_Sudoku_1k_T4.ipynb (new)
  • README.md (one extra line)

@mohabedalgani mohabedalgani changed the title Add Google Colab Sudoku 1 k demo (T4-compatible) Add Google Colab Sudoku 1k demo (T4-compatible) Aug 2, 2025
@imoneoi
Copy link
Contributor

imoneoi commented Aug 3, 2025

Thanks for the PR! 99.3% seems amazing, it's much higher than reported in our paper. Is that result from evaluate.py? I can't find the result of evaluation cell in the notebook.

@benman1
Copy link

benman1 commented Aug 16, 2025

This runs but I've never managed to get any correct examples in test in spite of trying. The model architecture is completely different. This is just a generic transformer model.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants