datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
sokoban_processed
Sokoban Processed Dataset
This folder contains a processed version of the Sokoban dataset derived from https://huggingface.co/datasets/Xiaofeng77/sokoban. The data is packaged for fast local loading and visual inspection.
Files: train.parquet (3,982 rows) and test.parquet (1,602 rows); corresponding PNGs live in images/.
Columns:
data_source: Source split/config from the original Hugging Face dataset (e.g., sokoban_6x6_1horizon).
prompt: Chat-style, multi-turn prompt used to elicit… See the full description on the dataset page: https://huggingface.co/datasets/VeraIsHere/sokoban_processed.sokoban_hard_3box_remap_prev
VisGym sokoban_hard_3box Remap-Prev
Solver-success Sokoban trajectories generated from the VisGym Sokoban environment.
HF repo: https://huggingface.co/datasets/novastar112/sokoban_hard_3box_remap_prev
task: sokoban_hard_3box
env_id: sokoban/hard
env_kwargs: {"action_mode": "move_only", "dim_room": [10, 10], "max_steps": 40, "num_boxes": 3, "num_gen_steps": 34, "success_mode": "stop_then_solved"}
action schema: move-only ('move', 1..4) plus ('stop', 'stop')
success semantics:… See the full description on the dataset page: https://huggingface.co/datasets/novastar112/sokoban_hard_3box_remap_prev.sokoban_easy_1box_remap_prev
VisGym sokoban_easy_1box Remap-Prev
Solver-success Sokoban trajectories generated from the VisGym Sokoban environment.
HF repo: https://huggingface.co/datasets/novastar112/sokoban_easy_1box_remap_prev
task: sokoban_easy_1box
env_id: sokoban/easy
env_kwargs: {"action_mode": "move_only", "dim_room": [7, 7], "max_steps": 20, "num_boxes": 1, "num_gen_steps": 23, "success_mode": "stop_then_solved"}
action schema: move-only ('move', 1..4) plus ('stop', 'stop')
success semantics:… See the full description on the dataset page: https://huggingface.co/datasets/novastar112/sokoban_easy_1box_remap_prev.sokoban_hard_remap_imagined_rollout_5000_cot_v3
Sokoban Hard Interleaved COT v3
Difficulty: hard.
This dataset is a v3 training view derived from two train sources:
5,000 imagined-rollout COT rows from the copied source trajectories.
5,000 disjoint final-stop COT rows from novastar111/s_v5_move_only_remap_prev / sokoban_hard.
Main changes from v1:
The selected imagined-rollout COT conversation is truncated at the COT/action assistant turn.
The assistant-side current/context image after Now I observe... is removed.
Final-stop… See the full description on the dataset page: https://huggingface.co/datasets/novastar112/sokoban_hard_remap_imagined_rollout_5000_cot_v3.sokoban_hard_5box_remap_prev
VisGym sokoban_hard_5box Remap-Prev
Solver-success Sokoban trajectories generated from the VisGym Sokoban environment.
HF repo: https://huggingface.co/datasets/novastar112/sokoban_hard_5box_remap_prev
task: sokoban_hard_5box
env_id: sokoban/hard
env_kwargs: {"action_mode": "move_only", "dim_room": [10, 10], "max_steps": 40, "num_boxes": 5, "num_gen_steps": 34, "success_mode": "stop_then_solved"}
action schema: move-only ('move', 1..4) plus ('stop', 'stop')
success semantics:… See the full description on the dataset page: https://huggingface.co/datasets/novastar112/sokoban_hard_5box_remap_prev.