The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
parts: struct<part0: struct<episodes: int64, actual_transitions: int64, wall_seconds: double, workers: int6 (... 506 chars omitted)
child 0, part0: struct<episodes: int64, actual_transitions: int64, wall_seconds: double, workers: int64, frames_sha2 (... 11 chars omitted)
child 0, episodes: int64
child 1, actual_transitions: int64
child 2, wall_seconds: double
child 3, workers: int64
child 4, frames_sha256: string
child 1, part1: struct<episodes: int64, actual_transitions: int64, wall_seconds: double, workers: int64, frames_sha2 (... 11 chars omitted)
child 0, episodes: int64
child 1, actual_transitions: int64
child 2, wall_seconds: double
child 3, workers: int64
child 4, frames_sha256: string
child 2, part2: struct<episodes: int64, actual_transitions: int64, wall_seconds: double, workers: int64, frames_sha2 (... 11 chars omitted)
child 0, episodes: int64
child 1, actual_transitions: int64
child 2, wall_seconds: double
child 3, workers: int64
child 4, frames_sha256: string
child 3, part3: struct<episodes: int64, actual_transitions: int64, wall_seconds: double, workers: int64, frames_sha2 (... 11 chars omitted)
child 0, episodes: int64
child 1, actual_transitions: int64
child 2, wall_seconds: double
child 3, workers: int64
child 4, frames_sha256: string
child 4, part4: struct<episodes: int64, actual_transitions: int64, wall_seconds: double, workers:
...
LOSE: int64
child 1, MAX_FRAMES: int64
child 4, forward_bias: struct<episodes: int64, transitions: int64, length_mean: double, length_median: double, length_min: (... 87 chars omitted)
child 0, episodes: int64
child 1, transitions: int64
child 2, length_mean: double
child 3, length_median: double
child 4, length_min: int64
child 5, length_max: int64
child 6, outcomes: struct<LOSE: int64, MAX_FRAMES: int64, WIN: int64>
child 0, LOSE: int64
child 1, MAX_FRAMES: int64
child 2, WIN: int64
by_split: struct<>
by_generator: struct<>
by_split_transitions: struct<train: int64, validation: int64, test: int64>
child 0, train: int64
child 1, validation: int64
child 2, test: int64
by_generator_episodes: struct<ge: int64, hopper: int64, notch: int64, notchParam: int64, notchParamRand: int64, patternCoun (... 86 chars omitted)
child 0, ge: int64
child 1, hopper: int64
child 2, notch: int64
child 3, notchParam: int64
child 4, notchParamRand: int64
child 5, patternCount: int64
child 6, patternOccur: int64
child 7, patternWeightCount: int64
child 8, original: int64
child 9, ore: int64
total: struct<episodes: int64, transitions: int64, death_episodes: int64, events: int64, wall_seconds_sum: (... 7 chars omitted)
child 0, episodes: int64
child 1, transitions: int64
child 2, death_episodes: int64
child 3, events: int64
child 4, wall_seconds_sum: double
total_transitions: int64
to
{'parts': {'part0': {'episodes': Value('int64'), 'actual_transitions': Value('int64'), 'wall_seconds': Value('float64')}, 'part1': {'episodes': Value('int64'), 'actual_transitions': Value('int64'), 'wall_seconds': Value('float64')}, 'part2': {'episodes': Value('int64'), 'actual_transitions': Value('int64'), 'wall_seconds': Value('float64')}, 'part3': {'episodes': Value('int64'), 'actual_transitions': Value('int64'), 'wall_seconds': Value('float64')}, 'part4': {'episodes': Value('int64'), 'actual_transitions': Value('int64'), 'wall_seconds': Value('float64')}}, 'total_transitions': Value('int64')}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 149, in get_rows_or_raise
return get_rows(
dataset=dataset,
...<4 lines>...
column_names=column_names,
)
File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
return func(*args, **kwargs)
File "/src/services/worker/src/worker/utils.py", line 129, in get_rows
rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
File "/src/services/worker/src/worker/utils.py", line 544, in safe_iter
yield from ds.decode(False) if ds.features else ds
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 3052, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2559, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2584, in _iter_arrow
for key, pa_table in self.ex_iterable._iter_arrow():
~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 558, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 440, in _iter_arrow
for key, pa_table in self.generate_tables_fn(**gen_kwags):
~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 361, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2381, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2309, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
parts: struct<part0: struct<episodes: int64, actual_transitions: int64, wall_seconds: double, workers: int6 (... 506 chars omitted)
child 0, part0: struct<episodes: int64, actual_transitions: int64, wall_seconds: double, workers: int64, frames_sha2 (... 11 chars omitted)
child 0, episodes: int64
child 1, actual_transitions: int64
child 2, wall_seconds: double
child 3, workers: int64
child 4, frames_sha256: string
child 1, part1: struct<episodes: int64, actual_transitions: int64, wall_seconds: double, workers: int64, frames_sha2 (... 11 chars omitted)
child 0, episodes: int64
child 1, actual_transitions: int64
child 2, wall_seconds: double
child 3, workers: int64
child 4, frames_sha256: string
child 2, part2: struct<episodes: int64, actual_transitions: int64, wall_seconds: double, workers: int64, frames_sha2 (... 11 chars omitted)
child 0, episodes: int64
child 1, actual_transitions: int64
child 2, wall_seconds: double
child 3, workers: int64
child 4, frames_sha256: string
child 3, part3: struct<episodes: int64, actual_transitions: int64, wall_seconds: double, workers: int64, frames_sha2 (... 11 chars omitted)
child 0, episodes: int64
child 1, actual_transitions: int64
child 2, wall_seconds: double
child 3, workers: int64
child 4, frames_sha256: string
child 4, part4: struct<episodes: int64, actual_transitions: int64, wall_seconds: double, workers:
...
LOSE: int64
child 1, MAX_FRAMES: int64
child 4, forward_bias: struct<episodes: int64, transitions: int64, length_mean: double, length_median: double, length_min: (... 87 chars omitted)
child 0, episodes: int64
child 1, transitions: int64
child 2, length_mean: double
child 3, length_median: double
child 4, length_min: int64
child 5, length_max: int64
child 6, outcomes: struct<LOSE: int64, MAX_FRAMES: int64, WIN: int64>
child 0, LOSE: int64
child 1, MAX_FRAMES: int64
child 2, WIN: int64
by_split: struct<>
by_generator: struct<>
by_split_transitions: struct<train: int64, validation: int64, test: int64>
child 0, train: int64
child 1, validation: int64
child 2, test: int64
by_generator_episodes: struct<ge: int64, hopper: int64, notch: int64, notchParam: int64, notchParamRand: int64, patternCoun (... 86 chars omitted)
child 0, ge: int64
child 1, hopper: int64
child 2, notch: int64
child 3, notchParam: int64
child 4, notchParamRand: int64
child 5, patternCount: int64
child 6, patternOccur: int64
child 7, patternWeightCount: int64
child 8, original: int64
child 9, ore: int64
total: struct<episodes: int64, transitions: int64, death_episodes: int64, events: int64, wall_seconds_sum: (... 7 chars omitted)
child 0, episodes: int64
child 1, transitions: int64
child 2, death_episodes: int64
child 3, events: int64
child 4, wall_seconds_sum: double
total_transitions: int64
to
{'parts': {'part0': {'episodes': Value('int64'), 'actual_transitions': Value('int64'), 'wall_seconds': Value('float64')}, 'part1': {'episodes': Value('int64'), 'actual_transitions': Value('int64'), 'wall_seconds': Value('float64')}, 'part2': {'episodes': Value('int64'), 'actual_transitions': Value('int64'), 'wall_seconds': Value('float64')}, 'part3': {'episodes': Value('int64'), 'actual_transitions': Value('int64'), 'wall_seconds': Value('float64')}, 'part4': {'episodes': Value('int64'), 'actual_transitions': Value('int64'), 'wall_seconds': Value('float64')}}, 'total_transitions': Value('int64')}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
world-model-qa Mario data
Data for training game world models on the Mario AI Framework engine (commit b0f01f224f868cdedf6a4b1f75ecc4c7a1f53545, engine unmodified). From the world-model-qa project of Miraiyochi, inc.
| Path | Contents | Frames | Original | Stored |
|---|---|---|---|---|
m1_full_v1/part{0..4}/ |
M1 corpus. 2,500 episodes: 500 levels × 5 policies (Robin Baumgarten A*, A* without run, A* with ε=0.2, random hold, forward-biased random), seed 42, up to 1,500 updates each. The 500 levels are 60 each from 8 generators (ge, hopper, notch, notchParam, notchParamRand, patternCount, patternOccur, patternWeightCount; 50 train / 5 validation / 5 test each) plus original (15) and ore (5) as test. Collected 2026-09-19. The 5 parts are disjoint level subsets (levels[k::5]), each a complete corpus. |
681,047 | 134.50 GB | 218 MB |
m1b_human_play/ (corpus_PI_001, corpus_PI_002) |
M1b human play. One human player on keyboard at 30 Hz, same corpus format plus human_metadata/ (per-tick timing, player settings), events_derived.jsonl, and per-tick value scores (value_PI_00*.npy). |
31,919 | 6.31 GB | 22 MB |
m2_dinov3_latents/latents_rev2_machine_only/ |
DINOv3 features of every M1 frame, fp16 memmap (681047, 1024) |
– | 1.39 GB | 1.39 GB |
m2_dinov3_latents/latents/ |
Same for M1 then M1b frames, fp16 memmap (712966, 1024) |
– | 1.46 GB | 1.46 GB |
Total: 143.67 GB uncompressed, 3.09 GB stored.
Decompressing
Frame files (*.u8) are stored as zstd level 19 with a 128 MiB window (--long=27). Before upload, every compressed file was decompressed on the source machine and its SHA-256 checked against the original.
zstd -d --long=27 m1_full_v1/part0/frames.u8.zst # -> frames.u8
sha256sum -c SHA256SUMS.original --ignore-missing # checks the decompressed files
import numpy as np
frames = np.memmap("m1_full_v1/part0/frames.u8", np.uint8, "r").reshape(-1, 256, 256, 3)
actions = np.load("m1_full_v1/part0/actions.npy")
SHA256SUMS lists the stored files, SHA256SUMS.original the uncompressed ones.
Corpus format (per part / per human corpus)
One row t is the frame before the update o_t, the action applied a_t, and the raw state after the update s_{t+1}. Each row is exactly one world.update. There is no frame skipping.
| File | Format |
|---|---|
frames.u8 |
Raw RGB with no header, shape (T, 256, 256, 3) uint8. Includes the initial frame. The bottom tile row is at pixel rows 240–255. |
terminal_frames.u8 |
Frame after the last update of each episode, shape (episodes, 256, 256, 3) |
actions.npy |
(T,) int64, 0–31. Bits: LEFT=bit0, RIGHT=bit1, DOWN=bit2, SPEED=bit3, JUMP=bit4 (RIGHT+SPEED+JUMP = 26) |
episode_starts.npy, episode_ends.npy |
int64, half-open [start, end), sorted, no gaps |
signals.npy |
(T, 25) float32, raw state after the update. Columns: x, y, xa, ya, onGround, mayJump, mode, gameStatus, remainingTime, coins, kills, completion, lives, currentTick, cameraX, cameraY, isLarge, isFire, pauseTimer, hurts, deaths, wins, update_index, exit_x, max_x. Definitions are in signals_manifest.json. |
rewards.npy |
(T, 6) float32 channels: d_x, death, win, coin, kill, hurt. Definitions are in rewards_manifest.json. |
episodes.jsonl |
Per-episode index: level, split, content hash, policy and parameters, seed, outcome, length, initial and final state |
events.jsonl |
Every engine event: episode_index, update_index, eventType, eventParam, marioState, marioX, marioY, time |
level_split.json |
Content hash and train/validation/test split per level |
build_manifest.json, extraction_manifest.json |
JDK version, hashes of engine, policy and image sources, frame SHA-256, counts |
gameStatus: RUNNING=0, WIN=1, LOSE=2, TIME_OUT=3. mode: small=0, large=1, fire=2. Recording is nominally 30 fps; the engine timer drops 30 ms per normal update.
The engine is deterministic. Replaying the recorded actions through the engine reproduced the recorded frames bit for bit in the project's checks.
DINOv3 latents
facebook/dinov3-vitl16-pretrain-lvd1689m (ViT-L/16) at 224×224, mean of patch tokens → LayerNorm → 1024 dims, then standardized per dimension with statistics from the M1 train split only (mean.npy, std.npy). latents_manifest.json records the model revision, preprocessing and the SHA-256 of the source frames. These fp16 files are the final standardized latents. The raw fp32 features from before standardization were not uploaded.
Notes
The frames are rendered by the Mario AI Framework, whose sprites depict Nintendo's Super Mario Bros. characters.
- Downloads last month
- 281