Dataset Viewer
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
format: string
ladder: string
pool: string
created: struct<corpus_config: struct<split: string, seed: int64, dataset: string>, utc: timestamp[s]>
child 0, corpus_config: struct<split: string, seed: int64, dataset: string>
child 0, split: string
child 1, seed: int64
child 2, dataset: string
child 1, utc: timestamp[s]
pairs: list<item: list<item: string>>
child 0, item: list<item: string>
child 0, item: string
nei_by_page: struct<Seth_MacFarlane: list<item: string>, The_World_According_to_Paris: list<item: string>, Joseph (... 554813 chars omitted)
child 0, Seth_MacFarlane: list<item: string>
child 0, item: string
child 1, The_World_According_to_Paris: list<item: string>
child 0, item: string
child 2, Joseph_Serrano: list<item: string>
child 0, item: string
child 3, L.A._Confidentiel: list<item: string>
child 0, item: string
child 4, Peloponnesian_War: list<item: string>
child 0, item: string
child 5, Kick-Ass_-LRB-film-RRB-: list<item: string>
child 0, item: string
child 6, Electoral_history_of_Jimmy_Carter: list<item: string>
child 0, item: string
child 7, Indiana: list<item: string>
child 0, item: string
child 8, This_Is_the_Life_-LRB-2008_film-RRB-: list<item: string>
child 0, item: string
child 9, Tim_Rice: list<item: string>
child 0, item: string
child 10, Anne_Bancroft: list<item: string>
child 0, item: string
child 11, Lincoln_Motor_Company: list<item:
...
913, Don_Simpson: list<item: list<item: string>>
child 0, item: list<item: string>
child 0, item: string
child 2914, Sunflower_-LRB-1970_film-RRB-: list<item: list<item: string>>
child 0, item: list<item: string>
child 0, item: string
child 2915, Shonen_Jump_-LRB-magazine-RRB-: list<item: list<item: string>>
child 0, item: list<item: string>
child 0, item: string
child 2916, Equidae: list<item: list<item: string>>
child 0, item: list<item: string>
child 0, item: string
child 2917, Adderall: list<item: list<item: string>>
child 0, item: list<item: string>
child 0, item: string
child 2918, Kung_Fu_Panda_3: list<item: list<item: string>>
child 0, item: list<item: string>
child 0, item: string
child 2919, Counterculture: list<item: list<item: string>>
child 0, item: list<item: string>
child 0, item: string
fillers: list<item: string>
child 0, item: string
pages: list<item: string>
child 0, item: string
runs: list<item: struct<book: string, sentences: list<item: string>>>
child 0, item: struct<book: string, sentences: list<item: string>>
child 0, book: string
child 1, sentences: list<item: string>
child 0, item: string
provenance: struct<corpus: string, books: int64, runs: int64, splitter: string, seed: int64>
child 0, corpus: string
child 1, books: int64
child 2, runs: int64
child 3, splitter: string
child 4, seed: int64
to
{'format': Value('string'), 'ladder': Value('string'), 'pool': Value('string'), 'created': {'corpus_config': {'hf_dataset': Value('string'), 'text_column': Value('string'), 'id_column': Value('null'), 'local_dir': Value('null'), 'max_books': Value('int64'), 'max_sentences_per_book': Value('int64'), 'min_run': Value('int64'), 'min_sentence_words': Value('int64'), 'splitter': Value('string'), 'seed': Value('int64')}, 'utc': Value('timestamp[s]')}, 'runs': List({'book': Value('string'), 'sentences': List(Value('string'))}), 'provenance': {'corpus': Value('string'), 'books': Value('int64'), 'runs': Value('int64'), 'splitter': Value('string'), 'seed': Value('int64')}}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 147, 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 127, 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 483, 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 2840, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, 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 536, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, 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 343, 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 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
format: string
ladder: string
pool: string
created: struct<corpus_config: struct<split: string, seed: int64, dataset: string>, utc: timestamp[s]>
child 0, corpus_config: struct<split: string, seed: int64, dataset: string>
child 0, split: string
child 1, seed: int64
child 2, dataset: string
child 1, utc: timestamp[s]
pairs: list<item: list<item: string>>
child 0, item: list<item: string>
child 0, item: string
nei_by_page: struct<Seth_MacFarlane: list<item: string>, The_World_According_to_Paris: list<item: string>, Joseph (... 554813 chars omitted)
child 0, Seth_MacFarlane: list<item: string>
child 0, item: string
child 1, The_World_According_to_Paris: list<item: string>
child 0, item: string
child 2, Joseph_Serrano: list<item: string>
child 0, item: string
child 3, L.A._Confidentiel: list<item: string>
child 0, item: string
child 4, Peloponnesian_War: list<item: string>
child 0, item: string
child 5, Kick-Ass_-LRB-film-RRB-: list<item: string>
child 0, item: string
child 6, Electoral_history_of_Jimmy_Carter: list<item: string>
child 0, item: string
child 7, Indiana: list<item: string>
child 0, item: string
child 8, This_Is_the_Life_-LRB-2008_film-RRB-: list<item: string>
child 0, item: string
child 9, Tim_Rice: list<item: string>
child 0, item: string
child 10, Anne_Bancroft: list<item: string>
child 0, item: string
child 11, Lincoln_Motor_Company: list<item:
...
913, Don_Simpson: list<item: list<item: string>>
child 0, item: list<item: string>
child 0, item: string
child 2914, Sunflower_-LRB-1970_film-RRB-: list<item: list<item: string>>
child 0, item: list<item: string>
child 0, item: string
child 2915, Shonen_Jump_-LRB-magazine-RRB-: list<item: list<item: string>>
child 0, item: list<item: string>
child 0, item: string
child 2916, Equidae: list<item: list<item: string>>
child 0, item: list<item: string>
child 0, item: string
child 2917, Adderall: list<item: list<item: string>>
child 0, item: list<item: string>
child 0, item: string
child 2918, Kung_Fu_Panda_3: list<item: list<item: string>>
child 0, item: list<item: string>
child 0, item: string
child 2919, Counterculture: list<item: list<item: string>>
child 0, item: list<item: string>
child 0, item: string
fillers: list<item: string>
child 0, item: string
pages: list<item: string>
child 0, item: string
runs: list<item: struct<book: string, sentences: list<item: string>>>
child 0, item: struct<book: string, sentences: list<item: string>>
child 0, book: string
child 1, sentences: list<item: string>
child 0, item: string
provenance: struct<corpus: string, books: int64, runs: int64, splitter: string, seed: int64>
child 0, corpus: string
child 1, books: int64
child 2, runs: int64
child 3, splitter: string
child 4, seed: int64
to
{'format': Value('string'), 'ladder': Value('string'), 'pool': Value('string'), 'created': {'corpus_config': {'hf_dataset': Value('string'), 'text_column': Value('string'), 'id_column': Value('null'), 'local_dir': Value('null'), 'max_books': Value('int64'), 'max_sentences_per_book': Value('int64'), 'min_run': Value('int64'), 'min_sentence_words': Value('int64'), 'splitter': Value('string'), 'seed': Value('int64')}, 'utc': Value('timestamp[s]')}, 'runs': List({'book': Value('string'), 'sentences': List(Value('string'))}), 'provenance': {'corpus': Value('string'), 'books': Value('int64'), 'runs': Value('int64'), 'splitter': Value('string'), 'seed': Value('int64')}}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
CTC seed pools
Serialized corpus pools for the CTC long-context suite's data generators
(allenai/OLMo-core, branch prasann/ctc,
pip package ctc/). Each file is the output of the one build step that needs heavy machinery — a
GPU cross-encoder, a pyserini/Lucene index, an LLM mining run, or a multi-gigabyte download —
captured once, so that anyone can build train and eval data at any context scale (2k to 10M+
tokens per example) on a bare pip install: no GPU, no Java, no API key.
git clone -b prasann/ctc https://github.com/allenai/OLMo-core && cd OLMo-core
pip install ./ctc
ctc-data build --task contradiction --pool auto --train 18000 --out DIR # ~4 min: 18k train
# + a nested 5-rung
# 500-example eval ladder
ctc-data build --task nq --pool auto --split eval --rungs 2k,32k --out DIR # seconds
ctc-data build --task contradiction --pool auto --split eval --rungs 64k,1m \
--eval-size 125 --allow-small-eval --out DIR # rungs extrapolate
# beyond the 32k table
--pool auto downloads <task>.seed.jsonl.gz from this repo (cached locally by
huggingface_hub; repeat builds are offline). A seeded build is the same build: the pool is
everything a generator reads, so identical (--seed, config) gives identical examples from the
live loader and from the file — asserted per ladder by the package's tests.
Format
Gzipped two-line JSONL: a header (format: ctc-seed-pool-v1, the ladder the pool was exported
for, provenance) and one payload line. Loading executes nothing but json.loads and whitelisted
dataclass constructors — no pickle. ctc-data build refuses a pool exported for a different
ladder, and ctc-data pool info FILE prints the header.
What each pool contains, and what it saved
| file | expensive part captured | notes |
|---|---|---|
contradiction.seed.jsonl.gz |
LLM-mined claim/contradiction pairs (60,342, recovered losslessly from the audited 20k train build) + PubMed filler abstracts | pairs are consumed, never reused: k=3 caps train at ~18k examples — pass --train 18000 |
redundancy.seed.jsonl.gz |
LLM-mined paraphrase pairs (4,477) + LLM-judged same-abstract hard negatives + fillers | supply-bounded like contradiction: ~1.3k train examples at the default k=3 |
nq.seed.jsonl.gz |
BM25 hard negatives from the 21M-passage wikipedia-dpr-100w Lucene index + GPU cross-encoder gold filter |
10% hard-negative regime, CE filter on; 9,093 distinct queries — a 20k train build reuses queries with fresh distractor draws and says so ("pool wraps") in its report |
hotpotqa.seed.jsonl.gz |
GPU cross-encoder ranking of the benchmark's distractors | bridge questions, 2 gold each; 25k queries |
rerank.seed.jsonl.gz |
MS MARCO mined hard negatives + a cross-encoder score for every document (25k queries) | the graded-ordering reference. Cannot wrap: fill is pre-drawn and scored per query, so distinct examples need distinct queries |
fiqa.seed.jsonl.gz / scifact.seed.jsonl.gz |
BEIR corpus + locally-built Lucene index + CE margin filter | eval-only ladders; build refuses --split train |
outlier.seed.jsonl.gz |
full scan of the 21M-passage wiki100w index into an article pool (2.2 GB) | largest file; expect a slow first load |
outlier_review.seed.jsonl.gz |
Amazon-Reviews-2023 streaming sample | eval-only |
contra_fever.seed.jsonl.gz |
FEVER gold-evidence restructuring | eval-only |
oolong.seed.jsonl.gz |
OOLONG-synth pull + per-item token counts (Qwen3 tokenizer) | |
absence.seed.jsonl.gz / reorder.seed.jsonl.gz |
Project Gutenberg (~11 GB) + punkt sentence segmentation into prose runs / passage streams | |
grouping_labeled.seed.jsonl.gz |
OpenAlex compact projection (52k papers + 31k year-restricted eval fetch) | the ~300 GB works snapshot, pre-reduced |
qdmatch_nq.seed.jsonl.gz / qdmatch_hpqa.seed.jsonl.gz |
projected from the nq / hotpotqa pools above | |
xabsence.seed.jsonl.gz |
LLM-mined paraphrase-twin pool | ⚠ 659 pairs only — seeds small builds; large rungs need a bigger mining run |
The four pure-synthetic ladders (cycle, groups4, mathmatch, textgroups) need no pool —
they build from a seed integer alone.
Pools are corpus material only (no rendered prompts, no eval answers beyond the corpora's own annotations). Underlying sources carry their own licenses: PubMedQA, Natural Questions, HotpotQA, MS MARCO, BEIR (FiQA/SciFact), FEVER, Amazon-Reviews-2023, OOLONG-synth, Project Gutenberg, OpenAlex, Wikipedia (DPR 100-word split).
- Downloads last month
- 26