Hugging Face's logo

Datasets:
maianh511
/
vi_chart_dataset

License:
Dataset card Data Studio Files
xet
 Community
Dataset Preview
Duplicate
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The dataset generation failed
Error code:   DatasetGenerationError
Exception:    CastError
Message:      Couldn't cast
id: string
image: struct<bytes: binary, path: string>
  child 0, bytes: binary
  child 1, path: string
description: string
conversations: list<item: struct<content: string, role: string>>
  child 0, item: struct<content: string, role: string>
      child 0, content: string
      child 1, role: string
groundtruth: string
-- schema metadata --
huggingface: '{"info": {"features": {"id": {"dtype": "string", "_type": "' + 265
to
{'indices': Value('uint64')}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1827, in _prepare_split_single
                  for key, table in generator:
                                    ^^^^^^^^^
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 613, in wrapped
                  for item in generator(*args, **kwargs):
                              ~~~~~~~~~^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/arrow/arrow.py", line 75, in _generate_tables
                  yield Key(file_idx, batch_idx), self._cast_table(pa_table)
                                                  ~~~~~~~~~~~~~~~~^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/arrow/arrow.py", line 54, 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 2306, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              id: string
              image: struct<bytes: binary, path: string>
                child 0, bytes: binary
                child 1, path: string
              description: string
              conversations: list<item: struct<content: string, role: string>>
                child 0, item: struct<content: string, role: string>
                    child 0, content: string
                    child 1, role: string
              groundtruth: string
              -- schema metadata --
              huggingface: '{"info": {"features": {"id": {"dtype": "string", "_type": "' + 265
              to
              {'indices': Value('uint64')}
              because column names don't match
              
              The above exception was the direct cause of the following exception:
              
              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 1880, in _prepare_split_single
                  raise DatasetGenerationError("An error occurred while generating the dataset") from e
              datasets.exceptions.DatasetGenerationError: An error occurred while generating the dataset

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.

indices
uint64
45,018
5,125
35,300
55,384
95,571
28,312
193
50,216
53,311
61,242
46,671
55,229
42,531
14,957
92,070
26,477
93,238
39,011
4,744
74,549
49,062
78,377
27,325
45,674
88,064
52,910
23,860
98,295
77,768
10,655
77,753
77,253
86,641
14,415
61,671
20,144
8,152
30,884
61,526
44,885
27,131
8,388
53,451
9,990
97,251
9,591
39,002
26,962
77,609
32,244
34,440
62,534
28,789
13,372
14,972
3,570
43,845
62,763
51,881
93,762
2,804
70,528
10,493
12,100
19,178
53,457
19,021
68,729
31,592
42,375
37,143
50,679
43,453
73,079
81,195
72,798
95,107
19,869
56,386
53,382
68,882
15,331
37,385
65,299
53,517
33,870
80,763
2,475
41,889
46,485
19,653
24,479
95,899
53,302
51,252
27,077
17,897
98,393
74,590
65,106
End of preview.

Dataset

The training data for the reasoning engine is built from 2 sources, combining structured chart data with natural Vietnamese language.

1. Viet-Chart-VQA (public benchmark)

A Vietnamese chart-based VQA benchmark covering 3 core chart types: bar charts (horizontal/vertical), line charts, and pie charts.

  • Image & metadata: high-quality chart images with dimensions and bounding box (plot-bb) info.
  • Natural language description: each chart paired with a Vietnamese description (axes, units, key trends).
  • Multi-turn conversation: questions (ranging from simple lookups to complex comparisons) with ground-truth answers in Vietnamese.
  • Structured ground truth: markdown/data-series tables mapping visual coordinates to logical values.

Limitation: chart labels/annotations are mostly in English; only the QA pairs are in Vietnamese.

2. Self-Built Vietnamese Chart Dataset (custom-built)

A supplementary dataset of native Vietnamese chart images (labels and annotations in Vietnamese) paired with Vietnamese questions, addressing the limitation above.

Each image is annotated with 5 QA pairs, covering 3 levels of information processing:

Level Description
Data Retrieval Extracting specific, atomic data points directly from the chart
Global Overview Synthesizing the chart's overall content/purpose
Data Reasoning Math/logical comparisons based on the chart's data
  • Answers are written as complete, grammatically correct Vietnamese narratives, not just numbers.

3. Data Preparation

  1. Standardization: unify the Q/A labeling format in the Self-Built dataset (originally hand-curated from heterogeneous sources).
  2. Sample splitting: samples with multiple QA pairs are split into independent single-QA samples, each keeping the original chart image.
  3. Sampling: randomly select 30,000 samples from the Viet-Chart-VQA train split (to prevent it from dominating the training signal).
  4. Train/test split: from Self-Built, reserve 200 samples for evaluation, the rest used for training.
  5. Resize: all images resized to 448×448 to match the vision encoder's input resolution.
  6. Export: each sample exported as an image + structured annotation pair, with question/answer explicitly identified.

Summary:

Split Composition
Train 30,000 samples (Viet-Chart-VQA) + Self-Built train samples
Test Original test split (Viet-Chart-VQA) + 200 reserved samples (Self-Built)
Downloads last month
19
Total file size:
3.32 GB

Models trained or fine-tuned on maianh511/vi_chart_dataset