Initial commit: 1.1M Multi-Game LLM Decision Training Dataset (23 Games, Train/Test SFT splits)
5fae0c6 verified Download sources.csv from kaushik-harsh-99/boardgame-decision-models: direct link, hf CLI and curl.
- Browser
- Download file 5.27 kB
-
https://huggingface.co/datasets/kaushik-harsh-99/boardgame-decision-models/resolve/main/sources.csv
- Command line
-
hf download hf://datasets/kaushik-harsh-99/boardgame-decision-models/sources.csv
-
curl -L -o sources.csv https://huggingface.co/datasets/kaushik-harsh-99/boardgame-decision-models/resolve/main/sources.csv
5.27 kB
| game,source,huggingface_url,license,accepted_examples,label_quality,notes | |
| tic_tac_toe,Exact Minimax Solver (all reachable non-terminal positions),N/A (Exhaustive Minimax Generation),MIT / Public Domain,4520,Proven Optimal,"Covers all 4,520 reachable distinct non-terminal game states in Tic-Tac-Toe." | |
| rock_paper_scissors,Game-Theoretic Best Response against specified opponent distributions,N/A (Game-Theoretic Simulation),MIT / Public Domain,1000,Proven Optimal,"High-density sample of mathematical counter-play against empirical frequencies, Markov transitions, and WSLS patterns." | |
| sudoku,sapientinc/sudoku-extreme,https://huggingface.co/datasets/sapientinc/sudoku-extreme,Apache 2.0 / CC-BY-4.0,75000,Proven Optimal,Single-step placement decisions on 9x9 grids with unique solution ground truth. | |
| 2048,yethdev/2048-gameplay-dataset,https://huggingface.co/datasets/yethdev/2048-gameplay-dataset,MIT,75000,Expert Play,"Action trajectories from expert gameplay reaching tiles 2048, 4096, and 8192." | |
| connect_four,homerquan/boardgamebench-answer-only,https://huggingface.co/datasets/homerquan/boardgamebench-answer-only,Apache 2.0,50000,Engine-guided,Evaluated with alpha-beta search connect four solver. | |
| chess,bingbangboom/chess-puzzles-san (Lichess),https://huggingface.co/datasets/bingbangboom/chess-puzzles-san,CC0 (Public Domain),200000,Engine Optimal,"Stockfish tactical puzzle moves with elo rating >= 1400 (pins, forks, mates, skewers, sacrifices)." | |
| checkers,NikolaiZhdanov/historical-checkers-games,https://huggingface.co/datasets/NikolaiZhdanov/historical-checkers-games,CC-BY-4.0 / MIT,100000,Master Play,"Master draughts play focused on decisions with >= 2 legal actions, mandatory jumps, king promotions, and winning play." | |
| reversi,homerquan/boardgamebench-answer-only,https://huggingface.co/datasets/homerquan/boardgamebench-answer-only,Apache 2.0,50000,Engine-guided,Evaluated with Edax/minimax Reversi engine. | |
| minesweeper,CL-From-Nothing/minesweeper_offline_14K,https://huggingface.co/datasets/CL-From-Nothing/minesweeper_offline_14K,Apache 2.0 / MIT,32677,Proven Safe,Constraint-solved safe cell selection without hidden-information leakage. | |
| snake,BLeaves/real-time-reasoning + A* Survival Agent,https://huggingface.co/datasets/BLeaves/real-time-reasoning,MIT,60000,Expert Planning,Target-seeking collision-avoidance trajectory steps in dynamic grid environment. | |
| tetris,lamshiu/gui-world-model-data (expert tetris),https://huggingface.co/datasets/lamshiu/gui-world-model-data,MIT,41125,Expert Play,Expert placement actions recorded from high-scoring Tetris play. | |
| blackjack,Dynamic Programming Casino Basic Strategy,N/A (Exact DP Solver),MIT / Public Domain,2500,Proven Optimal,Mathematically optimal Basic Strategy decisions across diverse multi-deck table configurations. | |
| battleship,ljt019/battleship-singleturn-synthetic-20000,https://huggingface.co/datasets/ljt019/battleship-singleturn-synthetic-20000,Apache 2.0,33516,Probability Targeting,Optimal targeting actions derived from fleet probability-density calculations without ship position leakage. | |
| poker,RZ412/PokerBench,https://huggingface.co/datasets/RZ412/PokerBench,MIT / CC-BY-4.0,60000,GTO Solver,PioSolver game-theory optimal postflop decisions across deep-stack 6-max scenarios. | |
| solitaire,chayuto/klondike-llm-decisions (teacher-clean),https://huggingface.co/datasets/chayuto/klondike-llm-decisions,MIT,11999,Teacher / Expert,Verified legal transitions with face-down cards strictly masked. | |
| mastermind,whoisjones/mastermind_24,https://huggingface.co/datasets/whoisjones/mastermind_24,Apache 2.0,40000,Exact Constraint,Logically consistent code selection by exact elimination of contradicted possibilities. | |
| yahtzee,Exact Expected-Score Dynamic Programming Solver,N/A (Exact DP Solver),MIT / Public Domain,28293,Proven Optimal,Mathematically proven optimal dice-keep and category assignments maximizing EV. | |
| dots_and_boxes,homerquan/boardgamebench-answer-only,https://huggingface.co/datasets/homerquan/boardgamebench-answer-only,Apache 2.0,25000,Engine-guided,Minimax-guided edge choices for 3x3 dots and boxes. | |
| sliding_puzzle,entangledloops/slidingpuzzle,https://huggingface.co/datasets/entangledloops/slidingpuzzle,MIT,75000,Proven Optimal,Shortest-path BFS optimal tile slider actions stratified across solution depths 1-31. | |
| pacman,Grid-based A* Planning Agent,N/A (A* Planning Simulator),MIT / Public Domain,60000,A* Planning Agent,Optimal navigation decisions balancing pellet collection and ghost evasion. | |
| sokoban,yuruny/sokoban_state_Markovian_S6-6_B1-1_D6-10_SFT_dataset,https://huggingface.co/datasets/yuruny/sokoban_state_Markovian_S6-6_B1-1_D6-10_SFT_dataset,Apache 2.0 / MIT,27656,Expert Planning,Verified expert decision steps on 6x6 Sokoban grid layouts. | |
| nim,Bouton 1901 Exact Nim-Sum / Sprague-Grundy Theorem (Exhaustive Solver),N/A (Exhaustive Mathematical Generation),MIT / Public Domain,20646,Proven Optimal,Exhaustive 3-pile and 4-pile combinatorial positions with mathematically proven optimal XOR decisions. | |
| tower_of_hanoi,"Exact Backward BFS Shortest-Path Solver (3 to 9 disks, exhaustive)",N/A (Exhaustive Mathematical Generation),MIT / Public Domain,29504,Proven Optimal,Exhaustive 3 to 9 disk non-terminal states with exact minimum-distance moves. | |