Skip to content
arc25
arc25
Initializing search
GitHub
Business Understanding
Data Understanding
State of the art
Initial Plan
Modeling
Utils
Solution summary
arc25
GitHub
Business Understanding
Data Understanding
State of the art
Initial Plan
Modeling
Modeling
Iteration 1. Architects baseline
Iteration 2. Architects solution with 8 data splits
Iteration 3. Ideal test-time training setup
Iteration 4. First steps with code
Iteration 5. Test-time training with code. Hindsight Experience Replay (HER)
Iteration 6. Reinforcement learning
Iteration 7. Optimize TTT on the evaluation set
Iteration 8. Improve HER
Iteration 9. Improve training script
Iteration 10. Try to solve real ARC tasks
Iteration 11. Pretrain LoRA on new tasks
Iteration 12. Solve a few ARC tasks
Iteration 13. Reflections
Iteration 14. Optimize inference
Iteration 15. The path forward: Search & Learn
Iteration 16. Search with base models
Iteration 17. Increase search diversity
Iteration 19. Search with BARC
Iteration 20. Data augmentation with BARC
Iteration 21. Fix bug with data
Iteration 22. Test-time Training with BARC induction model
Iteration 23. All in with test-time training with BARC induction model
Iteration 24. Using RL to improve BARC induction model
Iteration 25. Debug parallel code execution
Iteration 26. Acquire more compute
Iteration 27. Improve search and learn
Iteration 28. Refine predictions
Iteration 29. Multi-gpu RL
Iteration 30. Solve RL Collapse
Iteration 31. How to improve from 20% to 100%?
Iteration 32. Analyze model predictions
Iteration 33. RL with BARC data
Iteration 34. Multi-turn RL
Iteration 35. FP16 vs BF16
Iteration n. Iteration_title
Utils
Utils
Challenge workflow
Markdown cheatsheet
Methodology
Solution summary
arc25
Guillermo Barbadillo's solution for
ARC25 challenge
👉
Solution summary
Back to top
