saad1926q/8-puzzle
8-puzzle / 3x3 sliding puzzle Fixed 3x3 / 8-puzzle evaluation data and teacher-rollout datasets for sliding-puzzle reasoning experiments. Configs rl: 35 unique run-1 training puzzles at optimal depths 2-6, excluding all 31 fixed evaluation boards. It uses the run_1 split and contains only board and optimal_length. eval: 31 fixed evaluation puzzles, with exactly one puzzle at every optimal distance from 1 through 31. sft-source: 200 fresh boards, balanced with 20… See the full description on the dataset page: https://huggingface.co/datasets/saad1926q/8-puzzle.
Rebalance run_1 toward depths 2-3
Fix invalid RL split name
Document RL run-1 dataset split
Add RL run-1 training puzzles
Remove legacy 62-puzzle benchmark
Use one evaluation puzzle per depth
Document 62-puzzle evaluation benchmark
Remove retired exhaustive evaluation config
Replace evaluation data with 62-puzzle benchmark
Remove SFT validation split
Document fresh puzzle-level SFT validation split
Use all SFT decisions for training and fresh puzzle validation
Order SFT columns prompt before completion
Fix duplicate SFT validation card entry
Document bounded-history SFT train and validation splits
Replace SFT config with bounded-history decision records
Use evaluation prompt for SFT inputs
Order SFT conversation columns
Add replay-verified SFT trajectories
Balance SFT source across depths 10 to 19
Use 200 balanced boards for SFT source
Reduce SFT source pool to 1,000 boards
Add depth-balanced SFT source boards
Remove retired GEPA dataset metadata
Delete retired GEPA prompt-optimization splits
Add GEPA prompt-optimization splits
Add GEPA prompt-optimization splits
Add GEPA prompt-optimization splits
Add GEPA prompt-optimization splits
Update eval dataset card for 30 puzzles
Reduce eval split to 30 puzzles
docs: describe exhaustive evaluation split
feat: add exhaustive 8-puzzle evaluation split
Remove stale directional eval shard
Document tile action eval schema
Upload dataset
Regenerate balanced 45-row eval split
Replace eval split with stratified 45-row Parquet set
Upload README.md with huggingface_hub
Upload data/eval/eval.jsonl with huggingface_hub
initial commit