Datasets:

abdukuzi45
/

amharic-bench-2026-heldout

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
id: string
benchmark: string
language: string
split: string
task: string
premise: string
hypothesis: string
label: int64
label_text: string
prompt: string
raw: struct<header: list<item: string>, line: string>
  child 0, header: list<item: string>
      child 0, item: string
  child 1, line: string
answer: string
answer_index: int64
choices: list<item: string>
  child 0, item: string
question: string
subject: string
to
{'id': Value('string'), 'benchmark': Value('string'), 'language': Value('string'), 'split': Value('string'), 'task': Value('string'), 'subject': Value('string'), 'question': Value('string'), 'choices': List(Value('string')), 'answer': Value('string'), 'answer_index': Value('int64'), 'prompt': Value('string'), 'raw': {'header': List(Value('string')), 'line': Value('string')}}
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
              benchmark: string
              language: string
              split: string
              task: string
              premise: string
              hypothesis: string
              label: int64
              label_text: string
              prompt: string
              raw: struct<header: list<item: string>, line: string>
                child 0, header: list<item: string>
                    child 0, item: string
                child 1, line: string
              answer: string
              answer_index: int64
              choices: list<item: string>
                child 0, item: string
              question: string
              subject: string
              to
              {'id': Value('string'), 'benchmark': Value('string'), 'language': Value('string'), 'split': Value('string'), 'task': Value('string'), 'subject': Value('string'), 'question': Value('string'), 'choices': List(Value('string')), 'answer': Value('string'), 'answer_index': Value('int64'), 'prompt': Value('string'), 'raw': {'header': List(Value('string')), 'line': Value('string')}}
              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.

Amharic benchmark suite, 2026 — held out from training

Evaluation data for Amharic language models, kept in a repository of its own so it can never be mixed into a training corpus by accident.

Historical contamination scan: am=CLEAN, en=CLEAN for the pre-repair suite. The corrected prompts have not been rescanned. See contamination_report.json.

benchmark	task	am records	en records
afrimmlu	multiple_choice	607	608
afrixnli	natural_language_inference	1048	1050
belebele	reading_comprehension	900	900
masakhanews	topic_classification	564	1420

Each file is JSONL: one object per line with the benchmark's native fields plus a prompt string and an answer_index where the task has a discrete answer.

Loading

from datasets import load_dataset
ds = load_dataset("abdukuzi45/amharic-bench-2026-heldout", data_files="data/belebele/am/all.jsonl", split="train")
print(ds[0]["prompt"])

Why contamination was re-measured

The source corpus build report states that Amharic Belebele/StoryQA exclusions were inherited, not certified, and that no MMLU decontamination exists. This suite therefore does not rely on that claim. Every file shipped here was probed against the packed training rows with a rolling 12-token hash, verified by exact token match. Results are in contamination_report.json.

Known limitation: the probe is per row, so a passage split across a packed row boundary is not covered.

Citation

afrimmlu — Adelani et al., 2024. AfroMMLU: A Multilingual Benchmark for MMLU. (masakhane/afrimmlu)
afrixnli — Adelani et al., 2024. AfriXNLI: Cross-lingual NLI for African languages. (masakhane/afrixnli)
belebele — Bandarkar et al., 2023. Belebele: A reading comprehension benchmark. (facebook/belebele)
masakhanews — Adelani et al., 2023. MasakhaNEWS: News topic classification. (masakhane/masakhanews)

Structural repair — 2026-11-08

45 malformed AfriMMLU choice lists were recovered from their preserved raw TSV using proper CSV decoding. Prompts were rebuilt without changing gold labels. One confirmed invalid arithmetic question (afrimmlu-am-test-0000) was removed from active test data and preserved with a reason in quarantine/afrimmlu_invalid.jsonl. Active Amharic AfriMMLU test now has 499 records. Other benchmark data files are unchanged.

See repair_report_20261108.json for original records and audit details. This is not a complete semantic-label audit, and the historical contamination scan was not rerun. Quarantine files are not evaluation inputs. The original eval_benchmarks.py is a legacy runner; its GPU/scoring issues are not fixed by this data repair.

Downloads last month: 147

Total file size:

20.6 MB

Amharic benchmark suite, 2026 — held out from training

Contents

Loading

Why contamination was re-measured

Citation

Structural repair — 2026-11-08