jbduran/bart-dataset-v3

BART Dataset v3

The final version of the BART pretraining corpus, focused on removing anything that betrays a post-1930 origin. This is our best vintage dataset yet.

Documents	146,031 (97.52% of v2)
Characters	102,798,688,961 (96.73% of v2)
Tokens	~23B (estimated)
Shards	473 (one per v2 shard, same basename)
Source	BART Dataset v2
Cutoff	1930
Schema	single string column text

Lineage — three cumulative filtering stages over the same corpus:

Institutional Books 1.0 → v1 → v2 → v3

How it was made

For our final phase, we decided to focus on the time period. Although we filtered by date at v1, leaks and bad labeling inevitably brought modern documents in to contaminate the vintage corpus — publication metadata says nothing about a 1990s reprint's front matter, or an OCR pass that swept in a modern colophon.

So we filtered through document boilerplate once more, focusing on footers missed by earlier passes. But the biggest part of this phase was the regex anachronism filter: a tiered banned words list that flags documents and removes them on too many hits.

Rather than trimming, we removed entire documents when the conditions were met. A single Tier 1 hit was enough to discard a document, while Tier 2 words required two occurrences. Tier 3 words never triggered removal on their own — they only counted when accompanied by at least one Tier 2 hit.

We used this same strategy later in midtraining, to keep all of our data as vintage as possible.

The tiered anachronism filter

A document is dropped only on strong evidence of post-1930 content. The tiering exists to avoid polysemy false positives — "compiler of this volume", a bee "drone", birdsong "twitter", the Black Hole of Calcutta.

Tier	Terms	Behaviour
Tier 1	141	Coined well after 1930 — one hit drops the document
Tier 2	262	Real anachronisms — need ≥2 distinct tier-2/3 hits, with ≥1 tier-2
Tier 3	43	Polysemous, has a pre-1931 sense — never drops alone, only corroborates
Strip-only	17	Reproduction/boilerplate tells — never drops, only logged

Drop rule: 1× tier1 OR (≥2 distinct tier2/tier3 with ≥1 tier2)

The list was built from a 500-term seed, then curated: a 165-entry allow-list removed 55 seed terms, 18 were added by hand, leaving 463 final terms plus 10 format-tell patterns.

	Documents
Input	149,745
Kept	146,031 (97.52%)
Removed	3,714 (2.48%)
Characters kept	96.73%

Stage 0.5 — footer and boilerplate stripping

Runs before the anachronism filter. A line-level pass removes reprint and OCR footer lines — URLs, "printed in the United States of America", "all rights reserved", photocopy and print-on-demand colophons, ISBN lines, bare page numbers, library stamps — writing the stripped corpus to stripped/. Whole books are kept; only footer lines are removed. Documents that would lose more than 30% of their lines are kept unstripped and flagged.

Shards stripped	473
Documents changed	20,671 / 149,745
Documents flagged (kept unstripped)	0
Footer lines removed	31,761
Characters removed	0.0008%

Footer patterns by frequency

Pattern	Lines removed
page_num_bare	18,433
modern_year_reserved	7,324
printed_usa_modern	1,977
pod_preservation_photocopy	1,237
url_http	716
pod_ansi_paper	619
pod_laser	474
pod_acid_free_archival	265
lib_ex_libris	263
copyright_post_1930	174
url_www	145
url_dotcom	88
isbn	29
repro_this_was_produced	16
pod_lightning	1

Top firing anachronism terms

Terms that most often flagged a document. Note how many are polysemous — which is exactly why the tiering and allow-list exist.

Term	Hits		Term	Hits
compiler	515		mcdonalds	115
hardware	461		css	105
nagasaki	442		black hole	104
gaming	413		hiroshima	102
mcdonald's	375		blog	101
cia	342		united nations	98
drone	238		dark matter	98
nato	204		php	86
satellite	203		fax	83
rna	172		great depression	71
twitter	159		nasa	71
new deal	150		maser	66
dna	145		iron curtain	65
dvd	145		9/11	63
c#	128
holocaust	117

Schema and usage

Single string column named text.

from datasets import load_dataset

ds = load_dataset("jbduran/bart-dataset-v3", split="train", streaming=True)
print(next(iter(ds))["text"][:500])

Artifacts:

• _banned/ — the tiered lists, allow-list, format tells, and build audit
• stripped/ — the footer-stripped corpus
• strip_stats/, strip_samples/ — footer-strip audit
• stats/ — per-shard filter counts
• hits/ — per-shard hit log (which terms fired on which documents)
• scripts/ — the pipeline
• cleaning_report_1930s.json — full run report

Citation

Ultimately derived from Institutional Books 1.0:

@misc{institutionalbooks2025,
  title  = {Institutional Books 1.0: A 242B Token Dataset from Harvard Library's
            Collections, Refined for Accuracy and Usability},
  year   = {2025},
  eprint = {2506.08300},
  archivePrefix = {arXiv}
}