Datasets:

ysharma
/

north-micro-pope-multilingual

ArXiv:

Tags:

ml-intern

Dataset card Files Files and versions

xet

Community

Dataset Viewer

Duplicate

Split (1)

train

The dataset viewer is not available for this split.

Cannot load the dataset split (in streaming mode) to extract the first rows.

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
id: string
lang: string
question: string
gt: string
raw: string
pred: string
object_present: bool
model: string
n_images: int64
data: string
metrics: struct<en: struct<n: int64, hallucination_rate: double, accuracy: double, precision: double, recall: (... 446 chars omitted)
  child 0, en: struct<n: int64, hallucination_rate: double, accuracy: double, precision: double, recall: double, f1 (... 29 chars omitted)
      child 0, n: int64
      child 1, hallucination_rate: double
      child 2, accuracy: double
      child 3, precision: double
      child 4, recall: double
      child 5, f1: double
      child 6, unparseable: int64
  child 1, hi: struct<n: int64, hallucination_rate: double, accuracy: double, precision: double, recall: double, f1 (... 29 chars omitted)
      child 0, n: int64
      child 1, hallucination_rate: double
      child 2, accuracy: double
      child 3, precision: double
      child 4, recall: double
      child 5, f1: double
      child 6, unparseable: int64
  child 2, ja: struct<n: int64, hallucination_rate: double, accuracy: double, precision: double, recall: double, f1 (... 29 chars omitted)
      child 0, n: int64
      child 1, hallucination_rate: double
      child 2, accuracy: double
      child 3, precision: double
      child 4, recall: double
      child 5, f1: double
      child 6, unparseable: int64
  child 3, ar: struct<n: int64, hallucination_rate: double, accuracy: double, precision: double, recall: double, f1 (... 29 chars omitted)
      child 0, n: int64
      child 1, hallucination_rate: double
      child 2, accuracy: double
      child 3, precision: double
      child 4, recall: double
      child 5, f1: double
      child 6, unparseable: int64
to
{'model': Value('string'), 'data': Value('string'), 'n_images': Value('int64'), 'metrics': {'en': {'n': Value('int64'), 'hallucination_rate': Value('float64'), 'accuracy': Value('float64'), 'precision': Value('float64'), 'recall': Value('float64'), 'f1': Value('float64'), 'unparseable': Value('int64')}, 'hi': {'n': Value('int64'), 'hallucination_rate': Value('float64'), 'accuracy': Value('float64'), 'precision': Value('float64'), 'recall': Value('float64'), 'f1': Value('float64'), 'unparseable': Value('int64')}, 'ja': {'n': Value('int64'), 'hallucination_rate': Value('float64'), 'accuracy': Value('float64'), 'precision': Value('float64'), 'recall': Value('float64'), 'f1': Value('float64'), 'unparseable': Value('int64')}, 'ar': {'n': Value('int64'), 'hallucination_rate': Value('float64'), 'accuracy': Value('float64'), 'precision': Value('float64'), 'recall': Value('float64'), 'f1': Value('float64'), 'unparseable': 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
              id: string
              lang: string
              question: string
              gt: string
              raw: string
              pred: string
              object_present: bool
              model: string
              n_images: int64
              data: string
              metrics: struct<en: struct<n: int64, hallucination_rate: double, accuracy: double, precision: double, recall: (... 446 chars omitted)
                child 0, en: struct<n: int64, hallucination_rate: double, accuracy: double, precision: double, recall: double, f1 (... 29 chars omitted)
                    child 0, n: int64
                    child 1, hallucination_rate: double
                    child 2, accuracy: double
                    child 3, precision: double
                    child 4, recall: double
                    child 5, f1: double
                    child 6, unparseable: int64
                child 1, hi: struct<n: int64, hallucination_rate: double, accuracy: double, precision: double, recall: double, f1 (... 29 chars omitted)
                    child 0, n: int64
                    child 1, hallucination_rate: double
                    child 2, accuracy: double
                    child 3, precision: double
                    child 4, recall: double
                    child 5, f1: double
                    child 6, unparseable: int64
                child 2, ja: struct<n: int64, hallucination_rate: double, accuracy: double, precision: double, recall: double, f1 (... 29 chars omitted)
                    child 0, n: int64
                    child 1, hallucination_rate: double
                    child 2, accuracy: double
                    child 3, precision: double
                    child 4, recall: double
                    child 5, f1: double
                    child 6, unparseable: int64
                child 3, ar: struct<n: int64, hallucination_rate: double, accuracy: double, precision: double, recall: double, f1 (... 29 chars omitted)
                    child 0, n: int64
                    child 1, hallucination_rate: double
                    child 2, accuracy: double
                    child 3, precision: double
                    child 4, recall: double
                    child 5, f1: double
                    child 6, unparseable: int64
              to
              {'model': Value('string'), 'data': Value('string'), 'n_images': Value('int64'), 'metrics': {'en': {'n': Value('int64'), 'hallucination_rate': Value('float64'), 'accuracy': Value('float64'), 'precision': Value('float64'), 'recall': Value('float64'), 'f1': Value('float64'), 'unparseable': Value('int64')}, 'hi': {'n': Value('int64'), 'hallucination_rate': Value('float64'), 'accuracy': Value('float64'), 'precision': Value('float64'), 'recall': Value('float64'), 'f1': Value('float64'), 'unparseable': Value('int64')}, 'ja': {'n': Value('int64'), 'hallucination_rate': Value('float64'), 'accuracy': Value('float64'), 'precision': Value('float64'), 'recall': Value('float64'), 'f1': Value('float64'), 'unparseable': Value('int64')}, 'ar': {'n': Value('int64'), 'hallucination_rate': Value('float64'), 'accuracy': Value('float64'), 'precision': Value('float64'), 'recall': Value('float64'), 'f1': Value('float64'), 'unparseable': Value('int64')}}}
              because column names don't match

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.

Multilingual POPE hallucination eval — CohereLabs/North-Micro-Vision-Instruct

POPE-style "Is there a X?" hallucination evaluation of CohereLabs/North-Micro-Vision-Instruct in English, Hindi, Japanese and Arabic, run on a T4 GPU.

Setup

Data: 150 (image, question, answer) tuples sampled from the POPE adversarial split (COCO images; 75 object-present "yes" + 75 object-absent "no").
Languages: the same 150 tuples in en/hi/ja/ar = 600 queries. Questions translated with facebook/nllb-200-distilled-600M (beam 1); note machine translation is a confound — a few translations are imperfect (e.g. ja "sink"→洗濯機 "washing machine", "vase"→バース).
Protocol: one greedy short answer per query (max_new_tokens=8), answers parsed as yes/no in the question language (0 unparseable of 600). fp16 on T4 (the T4 does not support bf16).
Hallucination rate = fraction of object-absent questions answered "yes". Eval of arXiv:2305.10355 (POPE).

Results (150 questions per language)

Language	Hallucination rate	Accuracy	Precision	Recall	F1
English	6.7%	88.0%	92.5%	82.7%	87.3%
Hindi	4.0%	84.0%	94.7%	72.0%	81.8%
Japanese	5.3%	77.3%	91.8%	60.0%	72.6%
Arabic	12.0%	82.7%	86.6%	77.3%	81.7%

Interpretation: Arabic shows the highest object-hallucination rate (12% of absent-object questions answered "yes"); Japanese has a low hallucination rate but a marked drop in recall (misses 40% of present objects), suggesting the translated questions are harder to affirm than to deny — consistent with the MT-quality confound above.

Files

metrics.json — per-language metrics
results.jsonl — all 600 raw answers with predictions
chart.png — hallucination-rate bar chart

Downloads last month: 61

Total file size:

151 kB

Paper for ysharma/north-micro-pope-multilingual

Evaluating Object Hallucination in Large Vision-Language Models

Paper • 2305.10355 • Published May 17, 2023