The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: TypeError
Message: Couldn't cast array of type
struct<admission_scope: string, decision: string, human_approved: bool, model_trained: bool>
to
{'accepted_for_training': Value('bool'), 'status': Value('string')}
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 147, 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 127, 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 483, 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 2840, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, 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 536, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, 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 343, 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 2378, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2312, in cast_table_to_schema
cast_array_to_feature(
~~~~~~~~~~~~~~~~~~~~~^
table[name] if name in table_column_names else pa.array([None] * len(table), type=schema.field(name).type),
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
feature,
^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1861, in wrapper
return pa.chunked_array([func(chunk, *args, **kwargs) for chunk in array.chunks])
~~~~^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2118, in cast_array_to_feature
casted_array_values = _c(array.values, feature.feature)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1863, in wrapper
return func(array, *args, **kwargs)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2068, in cast_array_to_feature
_c(array.field(name) if name in array_fields else null_array, subfeature)
~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1863, in wrapper
return func(array, *args, **kwargs)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2158, in cast_array_to_feature
raise TypeError(f"Couldn't cast array of type\n{_short_str(array.type)}\nto\n{_short_str(feature)}")
TypeError: Couldn't cast array of type
struct<admission_scope: string, decision: string, human_approved: bool, model_trained: bool>
to
{'accepted_for_training': Value('bool'), 'status': Value('string')}Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
Speedpainting reference expansion
This repository archives licensed reference photographs, explicit bounded-brush-v2 programs, canonical 600×600 node renders, and independent assistant reviews for the reference-to-painting project.
The September 14 pilot contains 16 new reference photographs. After one geometry-repair round, 15 teacher targets were admitted for coarse structure supervision; the orchard remains held. These are manually composed Astra teacher programs, not Qwen outputs. No new model was trained on this pilot at publication. Admission establishes useful subject layout, connected parts, counts, negative spaces and broad color/value organization. It does not establish finished aesthetic quality or generalization of a trained model.
The preceding automatic region fitter produced fragmented subjects. Those targets were rejected. The manually composed programs retain explicit object geometry and were executed with pinned p5.brush and p5 assets in an isolated Linux/SwiftShader renderer. Reviews bind the exact reference, brush program and rendered image hashes, and record limitations. They are assistant reviews, not human labels.
Archive layout
Each snapshots/<sha256>/manifest.json lists explicit artifact paths, SHA256 hashes, content-addressed chunk paths, source attribution, source license and review status. Read the status metadata: archived artifacts may be held or superseded and must not automatically enter training. The immutable snapshot identifier and exact repository revision identify a reproducible archive. chunks/<prefix>/<sha256> stores the original file bytes.
Only frozen training references are publicly archived here. Validation and sealed references are kept separately. No credentials or model checkpoints are included. Model checkpoints remain in CK0607/qwen3.8-27b-brush-painting.
Attribution and licenses
There is no blanket image license. Each source record carries its title, author attribution, source URL, original license URL, credit, usage terms, image hashes and preprocessing. References are aspect-fitted into 600×600 canvases with neutral padding; geometry teachers and renders are adaptations. Per-artifact licenses are recorded in each snapshot and preserve the corresponding source license, including CC BY-SA when applicable. Retain that attribution and the stated license when redistributing or adapting individual files.
Intended next step
Expand independent references and reviewed geometry teachers, prepare bounded multi-turn continuations and verified repair episodes, mix general text/chart/document retention data, and audit the actual training processor and loss masks before another 27B SFT run. Multiple variants of a photo are not counted as independent references. The pilot alone is not a launch-ready dataset and is not an RL reward benchmark.
Reviewed pilot snapshot
The latest reviewed pilot is frozen at revision 746eef013814dbc0e97895ee7e4141924e31ec96: 112-file manifest. The archive includes the held orchard with explicit hold status; do not train on all archived files indiscriminately.
Second reviewed batch
The second batch adds 12 independent training photographs. After a targeted correction round, 11 teachers are admitted for coarse structure; farmland remains held. Together with the first pilot, the accepted training manifest now covers 26 photographs from 26 creator groups. The first pilot's orchard also remains held. These are assistant-reviewed teacher targets; no new Qwen SFT has started on them.
Five of six revised targets passed independent review: the museum return, succulent center, teapot color layout and two butterfly wing corrections. Farmland improved but still has material crop-density and cloud-shape deficits. Code validity and relative improvement alone do not grant admission. Four separately held-out validation targets are evaluation-only and are excluded from this public archive and from gradient training.
Latest exact revision 40e39e056064779ac3e68f32d6a97eb33a0cd32c: 84-file reviewed batch02 manifest. Archived hold-status files remain excluded from SFT. Earlier snapshots preserve the prior versions and rejection evidence.
Example teacher targets
Trams
| Reference | Structure teacher |
|---|---|
 |
 |
Source: File:Mk Berlin Tram 6.jpg, Magadan; CC BY-SA 3.0. Reference resized/padded; painting is a simplified geometry adaptation under the same stated license.
Violins
| Reference | Structure teacher |
|---|---|
 |
 |
Source: File:Violin-Viola.jpg, User:Frinck51; CC BY-SA 3.0. Reference resized/padded; painting is a simplified geometry adaptation under the same stated license.
- Downloads last month
- 109