HF RL Explorer

Solve a ramp of Sokoban puzzles turn-by-turn through a terminal game server, advancing through as many…

Solve a ramp of Sokoban puzzles turn-by-turn through a terminal game server, advancing through as many…: a task in Long-Horizon Terminal-Bench (LHTB) (Harbor dataset). You must SOLVE a series of Sokoban puzzles, one after another, advancing through as many levels as possible. There are 155 levels…

The task

You must SOLVE a series of Sokoban puzzles, one after another, advancing through as many levels as possible. There are 155 levels (the Microban set) that get progressively harder; nobody is expected to finish them all, so just push as far as you can. You play turn by turn through the `sokoban` command-line tool.…

Part of IntelligenceLab/Long-Horizon-Terminal-Bench.