Dataset Preview
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The dataset generation failed because of a cast error
Error code: DatasetGenerationCastError
Exception: DatasetGenerationCastError
Message: An error occurred while generating the dataset
All the data files must have the same columns, but at some point there are 5 new columns ({'shape_out', 'shape_in', 'secs', 'turns', 'budget'}) and 9 missing columns ({'cell_acc', 'exact', 'via', 'n_tok', 'n_train', 'test_shape', 'cost', 'reply', 'cut'}).
This happened while the json dataset builder was generating data using
hf://datasets/PardesiMusafir/arc-minimal-finetune/runs/realdesc_probe.json (at revision 1e21b0c2aac0b5528f793083ebd82ac935efa131), ['hf://datasets/PardesiMusafir/arc-minimal-finetune@1e21b0c2aac0b5528f793083ebd82ac935efa131/runs/battery_solve.json', 'hf://datasets/PardesiMusafir/arc-minimal-finetune@1e21b0c2aac0b5528f793083ebd82ac935efa131/runs/realdesc_probe.json', 'hf://datasets/PardesiMusafir/arc-minimal-finetune@1e21b0c2aac0b5528f793083ebd82ac935efa131/runs/realdesc_scored.json'], ['hf://datasets/PardesiMusafir/arc-minimal-finetune@1e21b0c2aac0b5528f793083ebd82ac935efa131/runs/battery_solve.json', 'hf://datasets/PardesiMusafir/arc-minimal-finetune@1e21b0c2aac0b5528f793083ebd82ac935efa131/runs/realdesc_probe.json', 'hf://datasets/PardesiMusafir/arc-minimal-finetune@1e21b0c2aac0b5528f793083ebd82ac935efa131/runs/realdesc_scored.json']
Please either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1848, in _prepare_split_single
writer.write_table(table)
~~~~~~~~~~~~~~~~~~^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 784, in write_table
self._write_table(pa_table, writer_batch_size=writer_batch_size)
~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 795, in _write_table
pa_table = table_cast(pa_table, self._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 2306, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
task: string
shape_in: list<item: int64>
child 0, item: int64
shape_out: list<item: int64>
child 0, item: int64
turns: list<item: struct<turn: int64, status: string, n_tok: int64, cut: bool, grids: struct<in_grid: struc (... 132 chars omitted)
child 0, item: struct<turn: int64, status: string, n_tok: int64, cut: bool, grids: struct<in_grid: struct<exact: bo (... 120 chars omitted)
child 0, turn: int64
child 1, status: string
child 2, n_tok: int64
child 3, cut: bool
child 4, grids: struct<in_grid: struct<exact: bool, cell_acc: double>, out_grid: struct<exact: bool, cell_acc: doubl (... 3 chars omitted)
child 0, in_grid: struct<exact: bool, cell_acc: double>
child 0, exact: bool
child 1, cell_acc: double
child 1, out_grid: struct<exact: bool, cell_acc: double>
child 0, exact: bool
child 1, cell_acc: double
child 5, corrections: list<item: string>
child 0, item: string
child 6, reply: string
secs: double
budget: int64
to
{'task': Value('string'), 'cost': Value('int64'), 'via': Value('string'), 'n_tok': Value('int64'), 'cut': Value('bool'), 'exact': Value('bool'), 'cell_acc': Value('float64'), 'n_train': Value('int64'), 'test_shape': List(Value('int64')), 'reply': Value('string')}
because column names don't match
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
~~~~~~~~~~~~~~~~~~~~~~~~~^
builder, max_dataset_size_bytes=max_dataset_size_bytes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1694, in _prepare_split
for job_id, done, content in self._prepare_split_single(
~~~~~~~~~~~~~~~~~~~~~~~~~~^
gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
):
^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1850, in _prepare_split_single
raise DatasetGenerationCastError.from_cast_error(
...<4 lines>...
)
datasets.exceptions.DatasetGenerationCastError: An error occurred while generating the dataset
All the data files must have the same columns, but at some point there are 5 new columns ({'shape_out', 'shape_in', 'secs', 'turns', 'budget'}) and 9 missing columns ({'cell_acc', 'exact', 'via', 'n_tok', 'n_train', 'test_shape', 'cost', 'reply', 'cut'}).
This happened while the json dataset builder was generating data using
hf://datasets/PardesiMusafir/arc-minimal-finetune/runs/realdesc_probe.json (at revision 1e21b0c2aac0b5528f793083ebd82ac935efa131), ['hf://datasets/PardesiMusafir/arc-minimal-finetune@1e21b0c2aac0b5528f793083ebd82ac935efa131/runs/battery_solve.json', 'hf://datasets/PardesiMusafir/arc-minimal-finetune@1e21b0c2aac0b5528f793083ebd82ac935efa131/runs/realdesc_probe.json', 'hf://datasets/PardesiMusafir/arc-minimal-finetune@1e21b0c2aac0b5528f793083ebd82ac935efa131/runs/realdesc_scored.json'], ['hf://datasets/PardesiMusafir/arc-minimal-finetune@1e21b0c2aac0b5528f793083ebd82ac935efa131/runs/battery_solve.json', 'hf://datasets/PardesiMusafir/arc-minimal-finetune@1e21b0c2aac0b5528f793083ebd82ac935efa131/runs/realdesc_probe.json', 'hf://datasets/PardesiMusafir/arc-minimal-finetune@1e21b0c2aac0b5528f793083ebd82ac935efa131/runs/realdesc_scored.json']
Please either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)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.
task string | cost int64 | via string | n_tok int64 | cut bool | exact bool | cell_acc float64 | n_train int64 | test_shape list | reply string |
|---|---|---|---|---|---|---|---|---|---|
16de56c4 | 975 | no_code | 2,560 | true | false | 0 | 3 | [
9,
21
] | Looking at the demonstration pairs, I can identify the transformation rule:
1. **train[0]**: Input has scattered digits (1s, 9s, 6s, 8s, 3s) in different rows. Output shows these same digits but arranged in horizontal lines at the bottom of the grid. The pattern is: collect all non-zero digits from each row and place ... |
291dc1e1 | 1,180 | crashed:IndentationError | 2,560 | true | false | 0 | 4 | [
18,
8
] | To solve this ARC task, I need to understand the transformation rule from the demonstration pairs.
**Analyzing the Demonstrations:**
1. **Structure of Input Grids:**
* The input grids contain a "background" or "frame" made of `8`s (and sometimes `1`s or `2`s on the border).
* Inside this frame, there are... |
7491f3cf | 1,575 | crashed:SyntaxError | 2,560 | true | false | 0 | 4 | [
7,
25
] | To solve this ARC task, let's analyze the transformation rule from the demonstration pairs.
**Analysis of Demonstration Pairs:**
Each pair consists of an input grid and an output grid of the same shape. The grids contain a border of a single color (e.g., 5, 1, 6, 4) and two internal rectangular regions separated by a... |
65b59efc | 1,725 | crashed:SyntaxError | 2,560 | true | false | 0 | 3 | [
25,
25
] | To solve this ARC task, I need to understand the transformation rule from the demonstration pairs.
**Analysis of Demonstration Pairs:**
1. **Pair 0:**
* **Input:** Contains three distinct rectangular regions of non-zero digits (values 1, 4, 2) separated by a grid of 5s. There are also some scattered single dig... |
332f06d7 | 1,792 | bad_prediction | 2,560 | true | false | 0 | 4 | [
20,
20
] | To solve this ARC task, let's analyze the transformation rule from the demonstration pairs.
**Analysis of Demonstration Pairs:**
1. **Pair 0:**
* **Input:** Contains a large region of `1`s with some `3`s (background) and a small `2x2` block of `0`s at the bottom right of the `1` region.
* **Output:** The... |
88bcf3b4 | 1,810 | code | 2,560 | true | false | 0 | 5 | [
22,
22
] | Looking at the demonstration pairs, I can identify the pattern:
1. There's a background color (0 in first pair, 7 in second, 1 in third, 0 in fourth, 0 in fifth)
2. There's a vertical line of a specific color (4, 4, 2, 2, 8, 8, 2, 2, 5, 5, 9, 9, 3, 3, 3, 3)
3. There's a shape made of another color (8, 8, 3, 3, 7, 7, 9... |
21897d95 | 1,936 | crashed:SyntaxError | 2,560 | true | false | 0 | 4 | [
30,
30
] | Looking at the demonstration pairs, I can identify the pattern:
1. The input grids contain multiple rectangular regions of different colors (numbers)
2. The output grids replace each rectangular region with a single row of that region's color, positioned at the top of the grid
3. The background color (9 in the test ca... |
c7f57c3e | 2,080 | crashed:SyntaxError | 2,560 | true | false | 0 | 2 | [
28,
28
] | Looking at the training data, I need to understand the transformation rule.
Let me analyze the pairs:
**train[0][0] → train[0][1]:**
- Background is 4
- There are several patterns of digits (1, 2, 3, 8)
- Looking at the differences:
- Top-right pattern: 1,2,1 with 8 below becomes 1,2,1 with 2 below (8→2)
- Bottom... |
58490d8a | 2,097 | crashed:SyntaxError | 2,560 | true | false | 0 | 3 | [
9,
11
] | To solve this ARC task, let's analyze the transformation rule from the demonstration pairs.
**Analysis of Demonstration Pairs:**
1. **Pair 0:**
* **Input:** A large grid filled mostly with `3`s. There are several small patterns of other digits (`8`, `2`, `4`) scattered within the `3` background.
* **Outp... |
dfadab01 | 2,400 | crashed:SyntaxError | 2,560 | true | false | 0 | 4 | [
20,
20
] | To solve this ARC task, I need to understand the transformation rule from the demonstration pairs.
**Analyzing the Demonstrations:**
1. **Pair 0:**
* **Input:** Contains scattered single pixels of value `6` (forming small 2x2 blocks), `2` (single pixels), `5` (single pixels), and `3` (single pixels). The `6`s ... |
8e5c0c38 | 2,420 | crashed:ValueError | 2,560 | true | false | 0 | 2 | [
22,
22
] | Looking at the training data, I can see a pattern:
1. In `train[0]`: Input has shapes made of 1s and 6s, output has the same shapes but with 2s instead of 1s and 6s. The background is 8.
2. In `train[1]`: Input has shapes made of 8s and 4s on background 0, output has the same shapes but with 2s instead of 8s and 4s.
... |
da515329 | 2,444 | code | 565 | false | false | 0 | 3 | [
26,
26
] | Looking at the demonstration pairs, I can see a clear pattern:
1. The input grids contain a small pattern of 8s (usually a cross or plus shape with one extra pixel)
2. The output grids contain the same pattern but scaled up by a factor of 2 in both dimensions
3. Each 8 in the input becomes a 2x2 block of 8s in the out... |
88bcf3b4 | null | null | null | null | null | null | null | null | null |
65b59efc | null | null | null | null | null | null | null | null | null |
16de56c4 | null | null | null | null | null | null | null | null | null |
291dc1e1 | null | null | null | null | null | null | null | null | null |
332f06d7 | null | null | null | null | null | null | null | null | null |
da515329 | null | null | null | null | null | null | null | null | null |
7491f3cf | null | null | null | null | null | null | null | null | null |
21897d95 | null | null | null | null | null | null | null | null | null |
58490d8a | null | null | null | null | null | null | null | null | null |
c7f57c3e | null | null | null | null | null | null | null | null | null |
3e6067c3 | null | null | null | null | null | null | null | null | null |
dfadab01 | null | null | null | null | null | null | null | null | null |
edb79dae | null | null | null | null | null | null | null | null | null |
4c416de3 | null | null | null | null | null | null | null | null | null |
0934a4d8 | null | null | null | null | null | null | null | null | null |
7b80bb43 | null | null | null | null | null | null | null | null | null |
8e5c0c38 | null | null | null | null | null | null | null | null | null |
a47bf94d | null | null | null | null | null | null | null | null | null |
b5ca7ac4 | null | null | null | null | null | null | null | null | null |
9aaea919 | null | null | null | null | null | null | null | null | null |
0934a4d8 | null | null | 2,723 | true | null | null | null | null | null |
16de56c4 | null | null | 241 | false | null | null | null | null | null |
21897d95 | null | null | 587 | false | null | null | null | null | null |
291dc1e1 | null | null | 1,798 | false | null | null | null | null | null |
332f06d7 | null | null | 1,540 | false | null | null | null | null | null |
3e6067c3 | null | null | 2,516 | true | null | null | null | null | null |
4c416de3 | null | null | 2,516 | true | null | null | null | null | null |
58490d8a | null | null | 1,626 | false | null | null | null | null | null |
65b59efc | null | null | 2,171 | false | null | null | null | null | null |
7491f3cf | null | null | 1,633 | false | null | null | null | null | null |
7b80bb43 | null | null | 1,228 | false | null | null | null | null | null |
88bcf3b4 | null | null | 240 | false | null | null | null | null | null |
8e5c0c38 | null | null | 565 | false | null | null | null | null | null |
9aaea919 | null | null | 1,610 | false | null | null | null | null | null |
a47bf94d | null | null | 463 | false | null | null | null | null | null |
b5ca7ac4 | null | null | 2,560 | true | null | null | null | null | null |
c7f57c3e | null | null | 871 | false | null | null | null | null | null |
da515329 | null | null | 511 | false | null | null | null | null | null |
dfadab01 | null | null | 509 | false | null | null | null | null | null |
edb79dae | null | null | 2,516 | true | null | null | null | null | null |
YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
arc-minimal-finetune
Runs from the finetune / skill-battery track of arc_minimal (its own pod; sources in that
repo's scratch/finetunes/). Kept separate from arc-minimal-runs so the two tracks cannot mix.
Every run here is on real ARC-AGI-2 evaluation tasks — no synthetic item family — and every score is produced by plain-code grading, never by a model.
| file | probe | what it measures |
|---|---|---|
runs/realdesc_probe.json |
describe | raw per-task replies + status for pair-0 grids of 20 eval tasks |
runs/realdesc_scored.json |
describe | the same replies re-scored through the fixed harness (see the doc) |
runs/battery_solve.json |
solve | the full task: demos + test input -> the test output grid |
Read SKILL_BATTERY.md for the harness fixes, the numbers, and what they mean. Headline: describe
is median 94.3 % cell accuracy on real grids (its failures are mostly harness artifacts), while
solve is 0/12 with 11 of 12 turns cut at the cap — and those tokens are spent on prose
analysis, not code. The next move is to force the answer into code, not to train.
- Downloads last month
- 73