Dataset Viewer
Duplicate
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
text: string
record_ids: list<item: string>
  child 0, item: string
record_id_format: string
source_json_sha256: string
source_counts_by_image_folder: struct<coco: int64, vg: int64, gqa: int64, ocr_vqa: int64, textvqa: int64, text_only: int64>
  child 0, coco: int64
  child 1, vg: int64
  child 2, gqa: int64
  child 3, ocr_vqa: int64
  child 4, textvqa: int64
  child 5, text_only: int64
count: int64
source_indices: list<item: int64>
  child 0, item: int64
eligible_pool_records_sha256: string
to
{'count': Value('int64'), 'record_id_format': Value('string'), 'source_json_sha256': Value('string'), 'eligible_pool_records_sha256': Value('string'), 'source_counts_by_image_folder': {'coco': Value('int64'), 'vg': Value('int64'), 'gqa': Value('int64'), 'ocr_vqa': Value('int64'), 'textvqa': Value('int64'), 'text_only': Value('int64')}, 'record_ids': List(Value('string')), 'source_indices': List(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
              text: string
              record_ids: list<item: string>
                child 0, item: string
              record_id_format: string
              source_json_sha256: string
              source_counts_by_image_folder: struct<coco: int64, vg: int64, gqa: int64, ocr_vqa: int64, textvqa: int64, text_only: int64>
                child 0, coco: int64
                child 1, vg: int64
                child 2, gqa: int64
                child 3, ocr_vqa: int64
                child 4, textvqa: int64
                child 5, text_only: int64
              count: int64
              source_indices: list<item: int64>
                child 0, item: int64
              eligible_pool_records_sha256: string
              to
              {'count': Value('int64'), 'record_id_format': Value('string'), 'source_json_sha256': Value('string'), 'eligible_pool_records_sha256': Value('string'), 'source_counts_by_image_folder': {'coco': Value('int64'), 'vg': Value('int64'), 'gqa': Value('int64'), 'ocr_vqa': Value('int64'), 'textvqa': Value('int64'), 'text_only': Value('int64')}, 'record_ids': List(Value('string')), 'source_indices': List(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.

trimdata baseline selections

Exact training subsets used for the baseline checkpoints in https://huggingface.co/submissionpaper1234/trimdata-baseline-checkpoints (code: https://github.com/LZhang12345/trimdata). Each selection is a subset of the 660,763 eligible records of the LLaVA-1.5 instruction mixture llava_v1_5_mix665k.json.

  • pool/pool_record_ids.json — the eligible pool: record_ids (<0-based index in llava_v1_5_mix665k.json>:<LLaVA id>), source_indices, and counts by image folder (coco: 359,700, vg: 86,352, ocr_vqa: 80,000, gqa: 72,070, text_only: 40,688, textvqa: 21,953).
  • selections/<method>/<budget>/<selection-seed>/record_ids.json — the ordered training records as pool ids (every record matched by full-record hash; 0 unmatched); selection_meta.json — method, K, seed, per-source counts and the sha256 of the exact training input records.json.
  • reconstruct_records.py — rebuilds that records.json byte-for-byte from llava_v1_5_mix665k.json (liuhaotian/LLaVA-Instruct-150K, revision 9d451dc7629cfe0469f6ae4432b765cd603d5fcb, sha256 ce959ce6e23073ee1cd1a8a2ef1c633768c10d4174327b8b2dc7113b91af6cf8) and checks the sha256. The conversation text is not re-hosted here; it is identical to the upstream records.
  • selections_index.csv, SHA256SUMS.

Images are not included; obtain them from the official LLaVA-1.5 data instructions. Evaluation-bank questions are not included either (they come from the official LLaVA eval.zip; hashes in the code repository's docs/EVAL_INPUTS.md).

Attribution and license

Data: the selections are subsets of the LLaVA-1.5 instruction-tuning mixture llava_v1_5_mix665k.json (LLaVA-Instruct / LLaVA-1.5 data, released by the LLaVA authors under CC BY 4.0; the original authors also ask users to abide by the OpenAI terms of use for the GPT-generated portions). Component datasets (COCO, GQA, OCR-VQA, TextVQA, Visual Genome, ShareGPT text) keep their own licenses. No images are redistributed.

This dataset describes subsets (record identifiers) of the LLaVA-1.5 instruction data, released under CC BY 4.0, with attribution to the LLaVA authors (Liu et al., "Visual Instruction Tuning" and "Improved Baselines with Visual Instruction Tuning").

Selections

Dataset path Method Budget K sel seed records.json sha256
selections/coincide/b05/selfixed COINCIDE (adaptation; see notes) 5% 33,038 fixed 24abed8530b7f2b7e89ae351d134f9baf0551d59273e31a02be9d9e87cb669ed
selections/coincide/b10/sel1234 COINCIDE (adaptation; see notes) 10% 66,076 1234 36d75b04009fe9ce4c8bed6b5be4c7da698a0327e81e59b373a64aae352debdf
selections/coincide/b20/selfixed COINCIDE (adaptation; see notes) 20% 132,152 fixed b38b714c95267e403689849a0c6efd992a9e3274161d93248a78f07fe7e200fa
selections/datatailor/b05/sel1234 DataTailor (official code) 5% 33,032 1234 4ef48c0305a10150fcf33f1e491d975262a329a1e39423f2b5a47c31b14042db
selections/datatailor/b10/sel1234 DataTailor (official code) 10% 66,071 1234 43a5eaf05dff650dc85d0564923af97273b6284921b81bcaa8419f1f7a79b88c
selections/datatailor/b15/sel1234 DataTailor (official code) 15% 99,108 1234 06daec696605cfbb9133c343bfe7f9a932449ff1a6f083b0b803ade03850b9ee
selections/datatailor/b20/sel1234 DataTailor (official code) 20% 132,147 1234 9f48baf83bbdcf6bb005518d1fcc205cd130e478ac85d4bc0689cd025feea82b
selections/less_approx/b10/sel1234 LESS-approx (not faithful LESS) 10% 66,076 1234 23054b9f7f69bb63ec257549381dbcbac8e823a3ac7750a3c4b43c21771a03ce
selections/source_sqrt/b10/sel1234 Source-sqrt (source-balanced random) 10% 66,076 1234 ce593e22db4f5c753ed4bd592b4d7a4260bb05681ac2ec6f9c0aec7c3a2c23fd
selections/uniform/b05/sel1235 Uniform random (matched-K) 5% 33,032 1235 c0b81ac856e5a0575ccfbf7370d39e5e9ff67a9615e21f0d389d7f1e2871c811
selections/uniform/b10/sel1234 Uniform random 10% 66,076 1234 563b8572a69730c229b309dde77bb146e15d098caf1d26bec7d1a83b787d45cd
selections/uniform/b10/sel1235 Uniform random 10% 66,076 1235 4b2104f8e32ababc98728823beb21f3083fd669be4507bdb9f35b3a191c3b459
selections/uniform/b10/sel1236 Uniform random 10% 66,076 1236 5573f3bb85a062460f36d93a8972e15bc0e2066a0bfd63d7b6979a40bcac8ae0
selections/uniform/b15/sel1235 Uniform random (matched-K) 15% 99,108 1235 4ec139524195f8fbff3d6cb2f01337ea23cf484a26c10063e34d7590c2be59a1
selections/uniform/b20/sel1235 Uniform random (matched-K) 20% 132,147 1235 9fce3a449186af63cae5ba24f1c4fbc309933217f50b0f88d7b546afb79da51f
selections/uniform_scored_random/b10/sel1234 Uniform scored-random (control) 10% 66,076 1234 7ec32ab31dcea9557d93a369ec799d88ce4273c922059018e8100918a9580a7e

Method notes

  • uniform — uniform random whole-conversation sample (b10: selection seeds 1234/1235/1236; b05/b15/b20: seed 1235 with K matched to DataTailor's realised K).
  • source_sqrt — per-source quotas proportional to the square root of the source size, random within source.
  • coincide — COINCIDE adaptation (author feature route + clustering on the full eligible pool; deterministic).
  • datatailor — official DataTailor code (commit d6fa5eb6); realised K 66,071 at 10%.
  • less_approx — not faithful LESS (SGD count-sketch gradient features, 3×Uniform pre-filter, GQA-val/ScienceQA-val targets).
  • uniform_scored_random — control: uniform random sample produced through the scoring/materialisation pipeline.
Downloads last month
70