open-index/fineweb-2-nlp

FineWeb-2 NLP

23,235,231 sentences and 371,647,537 word tokens across 996 languages, extracted from 528,708 source documents (789.7 MB source data) in FineWeb-2. Every sentence, paragraph, word frequency, and n-gram frequency, split with language-aware segmentation and continuously updated.

---

What is this?

FineWeb-2 is HuggingFace's multilingual web text corpus. It contains approximately 5 billion documents totaling 20 TB of text, drawn from roughly 100 Common Crawl snapshots spanning 2013 to 2024, and covering 1,868 language-script pairs. It is the largest curated multilingual web corpus publicly available today.

Working directly with FineWeb-2 is challenging. The raw data is enormous, and common NLP tasks like sentence extraction, word frequency analysis, or n-gram computation require downloading and processing terabytes of parquet files. Most researchers need just one language, or just the sentences, or just the word frequencies. They should not have to process the entire corpus to get there.

FineWeb-2 NLP solves this by pre-segmenting every document in FineWeb-2 into four linguistically useful units:

Type	Rows	What you get
sentences	23,235,231	One row per sentence, with source document ID, URL, and position index
paragraphs	549,363	One row per paragraph, with sentence count per paragraph
words	20,542,963	Per-shard word frequency and document frequency tables
ngrams	475,285,790	Per-shard bigram through 5-gram frequency tables

Every row traces back to its source document through doc_id and doc_url fields, making it possible to navigate from any sentence or word back to the original web page. This traceability is important for research that needs to verify context, check for contamination, or build training sets with known provenance.

Why per-shard frequency tables?

Words and n-grams are computed per source shard rather than aggregated into a single global table for each language. This design choice is intentional: some languages in FineWeb-2 contain over 700 million documents, and building a single frequency table for that volume would require holding hundreds of millions of unique entries in memory simultaneously. By keeping frequencies per-shard, each output file stays small and self-contained.

Aggregation is straightforward. A single DuckDB query can combine all shards for a language in seconds:

-- Language-level word frequencies in one query
SELECT word, sum(frequency) as total_freq, sum(doc_frequency) as total_doc_freq
FROM 'hf://datasets/open-index/fineweb-2-nlp/data/words/lat_Latn/*.parquet'
GROUP BY word ORDER BY total_freq DESC LIMIT 100;

What is being released?

Four dataset configs, all stored as Zstandard-compressed Parquet files:

1. Sentences (config_name: sentences)

Column	Type	Description
sentence	string	The extracted sentence
doc_id	string	Source document UUID from FineWeb-2
doc_url	string	Original web page URL
position	int32	0-based sentence index within the document
language	string	ISO 639-3 language code (e.g. lat, vie, cmn)
language_script	string	ISO 15924 script (e.g. Latn, Hani, Cyrl)

2. Paragraphs (config_name: paragraphs)

Column	Type	Description
paragraph	string	The paragraph text
doc_id	string	Source document UUID
doc_url	string	Original web page URL
position	int32	0-based paragraph index within the document
language	string	ISO 639-3 code
language_script	string	ISO 15924 script
sentence_count	int32	Number of sentences detected in this paragraph

3. Words (config_name: words)

Column	Type	Description
word	string	Lowercased, NFC-normalized word
frequency	int64	Occurrence count within this shard
doc_frequency	int64	Documents containing this word (within shard)
language	string	ISO 639-3 code
language_script	string	ISO 15924 script

4. N-grams (config_name: ngrams)

Column	Type	Description
ngram	string	Space-joined n-gram (e.g. "of the", "in the world")
n	int32	N-gram size: 2 (bigram), 3 (trigram), 4, or 5
frequency	int64	Occurrence count within this shard
language	string	ISO 639-3 code
language_script	string	ISO 15924 script

Data organization

open-index/fineweb-2-nlp/
├── README.md
├── stats.csv
└── data/
    ├── sentences/
    │   ├── lat_Latn/
    │   │   └── 0000.parquet
    │   ├── vie_Latn/
    │   │   ├── 0000.parquet
    │   │   └── ...
    │   └── {lang_script}/
    │       └── {shard:04d}.parquet
    ├── paragraphs/
    │   └── {lang_script}/{shard:04d}.parquet
    ├── words/
    │   └── {lang_script}/{shard:04d}.parquet
    └── ngrams/
        └── {lang_script}/{shard:04d}.parquet

Each source FineWeb-2 shard maps to exactly one output file per type per language. Shard names are zero-padded four-digit integers (0000, 0001, ...) that match the source file ordering from HuggingFace.

Sentence distribution by language

non_Latn       ████████████████████████████████████████ 527,948
tuk_Cyrl       ████████████████ 221,123
alt_Cyrl       ███████████████ 205,540
qug_Latn       ██████████████ 193,962
tcz_Latn       ███████████ 152,266
gom_Latn       ███████████ 146,526
nbl_Latn       ██████████ 140,093
lua_Latn       ██████████ 138,946
mni_Latn       ██████████ 132,013
ssw_Latn       █████████ 127,536
kng_Latn       █████████ 126,936
mos_Latn       █████████ 126,123
mnw_Mymr       █████████ 120,477
pck_Latn       █████████ 119,867
tiv_Latn       ████████ 115,586
ron_Cyrl       ███████ 104,553
npi_Latn       ███████ 104,197
mdf_Cyrl       ███████ 103,765
nzi_Latn       ███████ 103,176
pam_Latn       ███████ 98,608
dak_Latn       ███████ 95,906
btx_Latn       ███████ 95,824
iso_Latn       ███████ 95,104
ory_Latn       ███████ 93,598
mar_Latn       ███████ 93,468
dag_Latn       ██████ 87,526
bci_Latn       ██████ 86,266
sgs_Latn       ██████ 84,996
chk_Latn       ██████ 84,764
lzh_Hani       ██████ 84,487

SQL to reproduce this chart

SELECT language || '_' || language_script as lang, count(*) as sentences
FROM 'hf://datasets/open-index/fineweb-2-nlp/data/sentences/**/*.parquet'
GROUP BY lang ORDER BY sentences DESC LIMIT 30;

Paragraph distribution by language

mos_Latn       ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ 3,396
szy_Latn       ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ 3,107
mdf_Cyrl       ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ 3,095
non_Latn       ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ 2,970
nah_Latn       ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ 2,823
glv_Latn       ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ 2,815
tok_Latn       ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ 2,807
sgs_Latn       ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ 2,753
gcf_Latn       ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ 2,688
npi_Latn       ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ 2,687
bjn_Latn       ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ 2,568
nbl_Latn       ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ 2,559
aaz_Latn       ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ 2,559
mnw_Mymr       ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ 2,553
acd_Latn       ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ 2,546
ach_Latn       ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ 2,546
sms_Latn       ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ 2,520
nzi_Latn       ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ 2,493
tcz_Latn       ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ 2,468
nak_Latn       ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ 2,433

SQL to reproduce this chart

SELECT language || '_' || language_script as lang, count(*) as paragraphs
FROM 'hf://datasets/open-index/fineweb-2-nlp/data/paragraphs/**/*.parquet'
GROUP BY lang ORDER BY paragraphs DESC LIMIT 20;

Splitting quality overview

ade_Latn       ░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 386.5 sent/doc
swg_Latn       ░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 302.5 sent/doc
tuk_Cyrl       ░░░░░░░░░░░░░░░░░░░░░░░░░░░ 267.7 sent/doc
dak_Latn       ░░░░░░░░░░░░░░░░░░░░░░░░ 232.8 sent/doc
non_Latn       ░░░░░░░░░░░░░░░░░░░░░░░ 226.5 sent/doc
pkb_Latn       ░░░░░░░░░░░░░░░░░░░░ 197.3 sent/doc
lem_Latn       ░░░░░░░░░░░░░░░░░░░ 186.6 sent/doc
wob_Latn       ░░░░░░░░░░░░░░░░░ 167.1 sent/doc
guh_Latn       ░░░░░░░░░░░░░░░ 154.4 sent/doc
lzh_Hani       ░░░░░░░░░░░░░░░ 151.4 sent/doc
rmn_Grek       ░░░░░░░░░░░░░░░ 150.7 sent/doc
esk_Latn       ░░░░░░░░░░░░░░ 144.9 sent/doc
quh_Latn       ░░░░░░░░░░░░░░ 136.2 sent/doc
txu_Latn       ░░░░░░░░░░░░ 120.4 sent/doc
byr_Latn       ░░░░░░░░░░░░ 116.9 sent/doc
ian_Latn       ░░░░░░░░░░░░ 116.6 sent/doc
yss_Latn       ░░░░░░░░░░░ 115.2 sent/doc
cbt_Latn       ░░░░░░░░░░░ 113.4 sent/doc
amx_Latn       ░░░░░░░░░░░ 111.6 sent/doc
nab_Latn       ░░░░░░░░░░░ 111.3 sent/doc

The chart above shows the average number of sentences extracted per source document for each language. This metric serves as a rough proxy for content quality and structural richness. Languages where the average is high tend to contain longer, well-structured articles with clear paragraph and sentence boundaries. Languages with lower averages typically have shorter source documents, or they use scripts and punctuation patterns where automatic sentence boundary detection is more difficult.

How to download and use this dataset

1. DuckDB (recommended for exploration)

DuckDB can query HuggingFace parquet files directly over HTTP without downloading anything to disk. This makes it the fastest way to explore the dataset.

-- Count sentences per language
SELECT language, language_script, count(*) as sentences
FROM 'hf://datasets/open-index/fineweb-2-nlp/data/sentences/**/*.parquet'
GROUP BY ALL ORDER BY sentences DESC;

-- Read Latin sentences
SELECT sentence, doc_url
FROM 'hf://datasets/open-index/fineweb-2-nlp/data/sentences/lat_Latn/*.parquet'
LIMIT 20;

-- Top 100 most frequent words in a language
SELECT word, frequency, doc_frequency
FROM 'hf://datasets/open-index/fineweb-2-nlp/data/words/vie_Latn/*.parquet'
ORDER BY frequency DESC LIMIT 100;

-- Most common bigrams in Latin
SELECT ngram, frequency
FROM 'hf://datasets/open-index/fineweb-2-nlp/data/ngrams/lat_Latn/*.parquet'
WHERE n = 2
ORDER BY frequency DESC LIMIT 50;

-- Average sentences per document per language
SELECT language_script,
       count(DISTINCT doc_id) as docs,
       count(*) as sentences,
       round(count(*) * 1.0 / count(DISTINCT doc_id), 1) as avg_sent_per_doc
FROM 'hf://datasets/open-index/fineweb-2-nlp/data/sentences/**/*.parquet'
GROUP BY language_script ORDER BY sentences DESC LIMIT 20;

-- Aggregate word frequencies across all shards
SELECT word, sum(frequency) as total_freq
FROM 'hf://datasets/open-index/fineweb-2-nlp/data/words/lat_Latn/*.parquet'
GROUP BY word ORDER BY total_freq DESC LIMIT 50;

-- Find sentences containing a specific word
SELECT sentence, doc_url
FROM 'hf://datasets/open-index/fineweb-2-nlp/data/sentences/lat_Latn/*.parquet'
WHERE sentence ILIKE '%roma%'
LIMIT 20;

2. Python (datasets library)

from datasets import load_dataset

# Stream all sentences (no full download needed)
ds = load_dataset("open-index/fineweb-2-nlp", "sentences", split="train", streaming=True)
for row in ds.take(10):
    print(f"[{row['language']}] {row['sentence'][:100]}")

# Load paragraphs for a specific language
ds = load_dataset("open-index/fineweb-2-nlp", "paragraphs", split="train", streaming=True)
lat_paras = (row for row in ds if row["language"] == "lat")

# Word frequencies
ds = load_dataset("open-index/fineweb-2-nlp", "words", split="train", streaming=True)
for row in ds.take(20):
    print(f"{row['word']:20s} freq={row['frequency']:>8,}  doc_freq={row['doc_frequency']:>6,}")

# N-gram analysis
ds = load_dataset("open-index/fineweb-2-nlp", "ngrams", split="train", streaming=True)
bigrams = (row for row in ds if row["n"] == 2)

3. huggingface_hub CLI

# Download all Latin sentences
huggingface-cli download open-index/fineweb-2-nlp --include "data/sentences/lat_Latn/*" --repo-type dataset

# Download Vietnamese words and ngrams
huggingface-cli download open-index/fineweb-2-nlp --include "data/words/vie_Latn/*" "data/ngrams/vie_Latn/*" --repo-type dataset

# Download everything for one language
huggingface-cli download open-index/fineweb-2-nlp --include "data/*/lat_Latn/*" --repo-type dataset

4. pandas + DuckDB

import duckdb

conn = duckdb.connect()

# Latin sentences as DataFrame
df = conn.sql("""
    SELECT sentence, doc_url, position
    FROM 'hf://datasets/open-index/fineweb-2-nlp/data/sentences/lat_Latn/*.parquet'
    LIMIT 1000
""").df()
print(f"Loaded {len(df):,} sentences")
print(df.head(10))

# Word frequency analysis
words_df = conn.sql("""
    SELECT word, sum(frequency) as total_freq
    FROM 'hf://datasets/open-index/fineweb-2-nlp/data/words/lat_Latn/*.parquet'
    GROUP BY word ORDER BY total_freq DESC LIMIT 200
""").df()
print(words_df)

Dataset statistics

Metric	Value
Total sentences	23,235,231
Total paragraphs	549,363
Total word tokens	371,647,537
Unique word entries (per-shard)	20,542,963
Total n-gram entries (per-shard)	475,285,790
Languages processed	996
Source documents	528,708
Source data processed	789.7 MB
Output parquet size	6.2 GB
Avg sentence length	104.9 chars
Avg paragraph length	4478.8 chars
Avg sentences per document	43.9
Avg paragraphs per document	1.0
Avg sentences per paragraph	42.3

Per-language breakdown

#	Language	Sentences	Paragraphs	Words	Avg Sent	Avg Para	Docs	Shards	Source	Output
1	non_Latn (non_Latn)	527,948	2,970	8,729,010	96.7	17369.0	2,331	1	21.9 MB	71.1 MB
2	Turkmen (tuk_Cyrl)	221,123	1,128	2,324,099	144.3	28472.7	826	1	8.6 MB	24.9 MB
3	alt_Cyrl (alt_Cyrl)	205,540	1,967	2,231,510	148.0	15570.0	1,881	1	7.6 MB	20.9 MB
4	qug_Latn (qug_Latn)	193,962	2,237	2,262,307	93.5	8190.1	2,231	1	6.6 MB	20.0 MB
5	tcz_Latn (tcz_Latn)	152,266	2,468	3,431,253	130.4	8104.5	2,459	1	7.8 MB	20.9 MB
6	gom_Latn (gom_Latn)	146,526	1,942	1,880,520	86.6	6606.7	1,668	1	5.5 MB	18.1 MB
7	nbl_Latn (nbl_Latn)	140,093	2,559	1,337,358	85.0	4706.4	2,519	1	4.8 MB	12.7 MB
8	lua_Latn (lua_Latn)	138,946	2,341	2,303,160	106.2	6359.2	2,332	1	5.1 MB	16.0 MB
9	mni_Latn (mni_Latn)	132,013	1,407	1,495,153	77.6	7378.1	1,395	1	4.2 MB	12.5 MB
10	Swati (ssw_Latn)	127,536	2,230	1,157,650	81.9	4741.7	2,115	1	4.0 MB	16.0 MB
11	kng_Latn (kng_Latn)	126,936	2,303	2,225,243	96.3	5363.5	2,198	1	4.0 MB	12.4 MB
12	mos_Latn (mos_Latn)	126,123	3,396	2,163,421	89.8	3372.7	2,240	1	4.4 MB	13.1 MB
13	mnw_Mymr (mnw_Mymr)	120,477	2,553	3,465,581	231.1	10945.0	2,340	1	5.9 MB	18.9 MB
14	pck_Latn (pck_Latn)	119,867	1,872	2,709,591	122.3	7892.2	1,871	1	5.8 MB	17.5 MB
15	tiv_Latn (tiv_Latn)	115,586	2,161	2,160,577	85.1	4602.0	2,139	1	3.6 MB	11.6 MB
16	Romanian (ron_Cyrl)	104,553	1,974	1,717,190	186.3	9917.3	1,906	1	5.4 MB	19.5 MB
17	npi_Latn (npi_Latn)	104,197	2,687	1,813,816	106.1	4151.9	2,476	1	4.2 MB	14.2 MB
18	mdf_Cyrl (mdf_Cyrl)	103,765	3,095	1,029,290	139.2	4700.5	1,783	1	3.9 MB	13.2 MB
19	nzi_Latn (nzi_Latn)	103,176	2,493	1,828,257	104.6	4371.2	2,493	1	3.7 MB	16.3 MB
20	pam_Latn (pam_Latn)	98,608	2,162	1,140,731	69.7	3221.8	2,005	1	2.9 MB	15.0 MB
21	dak_Latn (dak_Latn)	95,906	1,328	1,259,965	81.2	5938.5	412	1	1.9 MB	19.6 MB
22	btx_Latn (btx_Latn)	95,824	2,305	1,384,785	91.1	3828.6	2,294	1	3.3 MB	10.4 MB
23	iso_Latn (iso_Latn)	95,104	2,196	1,890,498	106.6	4658.4	2,186	1	3.4 MB	11.1 MB
24	ory_Latn (ory_Latn)	93,598	1,442	1,631,705	99.8	6543.5	1,319	1	3.3 MB	12.0 MB
25	Marathi (mar_Latn)	93,468	1,806	1,199,222	76.5	4010.5	1,757	1	3.1 MB	8.9 MB
26	dag_Latn (dag_Latn)	87,526	1,870	759,224	53.9	2567.0	1,035	1	1.7 MB	6.3 MB
27	bci_Latn (bci_Latn)	86,266	1,507	1,385,649	82.8	4794.4	1,503	1	2.3 MB	8.2 MB
28	sgs_Latn (sgs_Latn)	84,996	2,753	744,509	67.9	2127.1	2,382	1	2.8 MB	12.1 MB
29	chk_Latn (chk_Latn)	84,764	1,707	1,383,072	95.1	4770.8	1,685	1	2.9 MB	10.3 MB
30	lzh_Hani (lzh_Hani)	84,487	586	2,117,940	92.9	13440.9	558	1	4.0 MB	15.2 MB
31	tvl_Latn (tvl_Latn)	84,410	1,751	2,122,926	115.8	5630.9	1,737	1	3.1 MB	10.8 MB
32	tzh_Latn (tzh_Latn)	83,775	1,839	1,671,169	119.6	5492.4	1,814	1	3.4 MB	10.8 MB
33	hmo_Latn (hmo_Latn)	82,002	1,765	1,294,799	90.7	4259.8	1,750	1	2.0 MB	9.9 MB
34	bem_Latn (bem_Latn)	81,947	1,975	1,286,155	103.7	4342.9	1,787	1	3.2 MB	9.7 MB
35	rar_Latn (rar_Latn)	80,792	1,859	1,642,706	95.6	4197.6	1,818	1	2.6 MB	8.9 MB
36	toi_Latn (toi_Latn)	80,526	1,652	936,388	90.5	4460.7	1,649	1	2.7 MB	8.0 MB
37	Old English (ang_Latn)	76,156	2,213	910,971	74.5	2598.3	2,001	1	2.6 MB	15.9 MB
38	arn_Latn (arn_Latn)	76,026	2,039	1,123,438	97.2	3662.1	1,927	1	3.0 MB	9.7 MB
39	quz_Latn (quz_Latn)	75,347	1,855	761,653	84.5	3470.5	1,528	1	2.5 MB	8.2 MB
40	tuc_Latn (tuc_Latn)	74,425	1,362	879,671	65.9	3652.3	1,362	1	1.6 MB	11.4 MB
41	zai_Latn (zai_Latn)	74,271	1,720	1,046,360	86.7	3784.8	1,698	1	2.3 MB	11.4 MB
42	srm_Latn (srm_Latn)	74,187	1,937	1,321,052	80.1	3103.9	1,936	1	1.9 MB	7.3 MB
43	mps_Latn (mps_Latn)	74,142	965	1,268,651	92.7	7196.5	965	1	1.9 MB	6.4 MB
44	gcf_Latn (gcf_Latn)	73,747	2,688	1,059,342	80.3	2230.2	2,433	1	2.7 MB	13.3 MB
45	orv_Cyrl (orv_Cyrl)	73,447	1,458	1,632,538	226.1	11437.9	1,372	1	5.1 MB	22.8 MB
46	sms_Latn (sms_Latn)	72,503	2,520	740,057	106.0	3076.9	2,478	1	2.8 MB	10.1 MB
47	Manx (glv_Latn)	72,120	2,815	1,232,597	99.1	2563.2	2,462	1	3.1 MB	16.4 MB
48	bru_Latn (bru_Latn)	70,108	1,023	1,316,408	112.9	7808.0	1,021	1	2.5 MB	10.6 MB
49	nah_Latn (nah_Latn)	69,851	2,823	693,924	75.1	1880.9	2,522	1	1.9 MB	13.2 MB
50	ach_Latn (ach_Latn)	69,776	2,546	1,213,339	87.9	2436.0	2,522	1	2.4 MB	7.9 MB
51	syc_Syrc (syc_Syrc)	69,636	1,393	5,158,313	465.3	23311.1	1,307	1	9.1 MB	31.6 MB
52	kmb_Latn (kmb_Latn)	69,286	1,324	1,033,815	80.7	4276.5	1,306	1	2.0 MB	9.9 MB
53	awa_Deva (awa_Deva)	68,871	1,905	1,565,158	162.8	5917.8	1,902	1	2.9 MB	9.9 MB
54	umb_Latn (umb_Latn)	68,289	1,319	924,878	81.1	4251.9	1,315	1	2.0 MB	6.2 MB
55	Kalmyk (xal_Cyrl)	67,499	1,478	672,016	115.1	5301.9	1,424	1	2.3 MB	10.2 MB
56	byr_Latn (byr_Latn)	67,117	574	556,136	81.1	9601.0	574	1	1.4 MB	15.1 MB
57	bjn_Latn (bjn_Latn)	66,827	2,568	872,195	89.2	2346.2	2,323	1	2.5 MB	13.5 MB
58	ubu_Latn (ubu_Latn)	66,772	960	1,195,913	121.6	8526.1	960	1	2.3 MB	13.1 MB
59	hmr_Latn (hmr_Latn)	66,120	1,619	1,413,234	110.8	4564.9	1,352	1	2.9 MB	9.2 MB
60	kos_Latn (kos_Latn)	65,989	1,631	1,098,529	85.5	3498.6	1,622	1	2.1 MB	10.9 MB

How it works

Source (HuggingFaceFW/fineweb-2)       Pipeline                       Output (open-index/fineweb-2-nlp)
┌──────────────────────────┐     ┌───────────────────────┐     ┌───────────────────────────┐
│ data/{lang}/train/       │     │  1. Download shard    │     │ data/sentences/{lang}/    │
│   000_00000.parquet      │────▶│  2. Read 10K batches  │────▶│   0000.parquet            │
│   000_00001.parquet      │     │  3. Split:            │     │ data/paragraphs/{lang}/   │
│   ...                    │     │     · paragraphs      │     │   0000.parquet            │
│                          │     │     · sentences       │     │ data/words/{lang}/        │
│ ~5 billion docs          │     │     · words + freq    │     │   0000.parquet            │
│ 1,868 lang-script pairs  │     │     · ngrams + freq   │     │ data/ngrams/{lang}/       │
│ 20 TB total              │     │  4. Write parquet     │     │   0000.parquet            │
│                          │     │  5. Publish to HF     │     │ stats.csv                 │
│                          │     │  6. Delete local      │     │ README.md (auto-generated)│
└──────────────────────────┘     └───────────────────────┘     └───────────────────────────┘

Pipeline details

The pipeline processes FineWeb-2 one shard at a time to keep resource usage predictable and bounded. This is the core design principle: at no point does the pipeline need to hold more than one shard's worth of data in memory or on disk.

1. Download. A single source parquet shard is fetched from HuggingFace. Downloads are idempotent: if the file already exists locally with the correct size, it is skipped.

2. Read. The shard is streamed in batches of 10,000 rows using parquet-go. This keeps memory usage constant at roughly 20 MB regardless of shard size. Each batch of documents is distributed across parallel workers for splitting.

3. Split. Each worker processes its share of documents, extracting paragraphs, sentences, words, and n-grams. Workers maintain their own local frequency maps, which are merged after the batch completes. This avoids lock contention and keeps throughput high.

4. Write. Results are written as Zstandard-compressed Parquet files with 50,000 rows per row group. Zstandard provides excellent compression ratios on text data while remaining fast to decompress.

5. Publish. The output parquet files, along with an updated stats.csv and a regenerated README.md, are committed to HuggingFace in a single atomic operation. If the commit fails due to rate limiting or a transient server error, it is retried with exponential backoff.

6. Clean up. After a successful publish, both the source and output files are deleted from local disk. This prevents disk usage from growing over time and allows the pipeline to process thousands of languages on a machine with limited storage.

Resource budgets

Resource	Budget	How
Memory	~200 MB	10K-row read batches, frequency maps pruned at 1M entries
Disk	~10 GB peak	One shard at a time, deleted after successful publish
Network	Sequential	One download at a time, retry on rate limit

The pipeline is fully resumable. A stats.csv file tracks every completed shard, so re-running the pipeline after an interruption will automatically skip all previously finished work and continue from where it left off.

Splitting methodology

Sentence splitting

Sentence segmentation is one of the harder problems in multilingual NLP. There is no universal rule for where sentences begin and end: different languages use different punctuation conventions, and web text frequently breaks the conventions of any language.

Our approach uses a set of punctuation and casing heuristics tuned for web text across many scripts. The rules are designed to be conservative, preferring to keep text together rather than over-splitting. For short texts (under 500 characters), we use sentencex, a Wikimedia project that provides language-specific sentence boundary detection with knowledge of each language's abbreviation patterns and punctuation norms.

Rule	Example	Behavior
Period + space + uppercase	world. The	Split
Abbreviation + period	Mr. Smith	No split
Decimal number	3.14 is	No split
Single-letter initial	J. K. Rowling	No split
CJK fullstop	世界。今天	Always split
Devanagari danda	text। next	Always split
Exclamation/question	really! What	Split
Newline after 10+ chars	long text\nNext	Split

For CJK languages (Chinese, Japanese, Korean), individual Han characters, Hiragana, Katakana, and Hangul syllables are each treated as separate word tokens, reflecting the character-level structure of these writing systems. This means that a Chinese sentence like "今天天气很好" produces six word tokens rather than being treated as a single unsplittable string.

Word splitting

Word extraction follows a straightforward pipeline designed to produce clean, normalized tokens suitable for frequency analysis:

1. NFC normalization (Unicode canonical composition) to ensure that equivalent character sequences are represented identically
2. Lowercase conversion for case-insensitive frequency counting
3. Splitting on non-letter, non-digit boundaries, while preserving apostrophes and hyphens that appear mid-word (e.g. "don't", "well-known")
4. Stripping of leading and trailing punctuation
5. Filtering of empty strings and pure-punctuation tokens

Paragraph splitting

FineWeb-2's source text comes from HTML pages processed by trafilatura, a web content extraction library. In trafilatura's output, HTML <p> tags are represented as double newlines (\n\n). We use this convention to split text into paragraphs:

1. Split on sequences of two or more consecutive newlines
2. Trim leading and trailing whitespace from each paragraph
3. Discard fragments shorter than 20 characters, which typically correspond to navigation elements, single-word headers, or other structural debris from the original HTML

This simple approach works well in practice because trafilatura has already done the hard work of extracting meaningful content blocks from the HTML.

N-gram extraction

N-grams are extracted by sliding a window of size n over the word token sequence for each document. We compute bigrams (n=2), trigrams (n=3), 4-grams, and 5-grams.

N	Name	Example from "the quick brown fox"
2	Bigram	"the quick", "quick brown", "brown fox"
3	Trigram	"the quick brown", "quick brown fox"
4	4-gram	"the quick brown fox"
5	5-gram	(needs 5+ words)

To keep memory usage bounded, per-shard frequency maps are pruned when they exceed 1 million unique entries. During pruning, entries with a frequency of 1 are evicted first. This means that very rare n-grams in large shards may be undercounted, but the most frequent and analytically useful n-grams are preserved accurately.

Dataset card

Dataset summary

FineWeb-2 NLP provides pre-segmented versions of HuggingFace's FineWeb-2 dataset. Each of the approximately 5 billion source documents has been split into sentences, paragraphs, words, and n-grams using language-aware processing. The four resulting datasets share document IDs, so researchers can cross-reference between them: look up which sentences appear in a document, check the word frequencies for that language, or find which n-grams co-occur with a particular sentence.

The primary goal is to lower the barrier to multilingual NLP research. Instead of downloading and processing 20 TB of raw text, researchers can query exactly the slice they need, whether that is all sentences in Latin, word frequencies in Vietnamese, or bigram distributions across every language in the corpus.

Data instances

Sentence:

{
  "sentence": "Gallia est omnis divisa in partes tres.",
  "doc_id": "f7ef49fc-6899-4d56-aaa7-bea5924802f3",
  "doc_url": "https://example.com/caesar",
  "position": 0,
  "language": "lat",
  "language_script": "Latn"
}

Word:

{
  "word": "est",
  "frequency": 847,
  "doc_frequency": 412,
  "language": "lat",
  "language_script": "Latn"
}

N-gram:

{
  "ngram": "in partes",
  "n": 2,
  "frequency": 23,
  "language": "lat",
  "language_script": "Latn"
}

Curation rationale

Sentence-level and word-level datasets are foundational for many areas of NLP research. They are used to train sentence embeddings, build and evaluate language models, study word frequency distributions and Zipf's law across languages, analyze collocations and phrasal patterns, and benchmark multilingual NLP tools. Having these units pre-extracted and ready to query saves researchers significant time and computational resources, and makes it practical to work with languages that might otherwise be overlooked due to the effort required to process the raw data.

Source data

All text originates from FineWeb-2 (DOI: 10.57967/hf/3744). FineWeb-2 was constructed by extracting text from approximately 100 Common Crawl snapshots covering 2013 through 2024. The extraction pipeline includes text extraction via trafilatura, language identification using GlotLID, MinHash deduplication to remove near-duplicate documents, and adaptive quality filtering to remove low-quality content. We do not apply any additional filtering or deduplication beyond what FineWeb-2 provides.

Considerations for using the data

There are several important limitations to keep in mind when working with this dataset:

Low-resource language coverage. Many of the smaller languages in FineWeb-2 consist primarily of Bible translations, Wikipedia mirrors, and religious texts. The FineWeb-2 authors note that over 70% of language-script pairs have more than 50% of their content from such sources. Word frequencies and n-gram distributions for these languages will reflect this narrow domain rather than general language use.

Sentence splitting accuracy. The quality of sentence segmentation varies by language and script. Latin-script and CJK languages tend to produce the most accurate results, because their punctuation conventions are well-understood and widely standardized. Languages with less common scripts, or languages that use minimal punctuation, may have lower splitting accuracy.

Vietnamese word boundaries. Vietnamese is written with spaces between syllables rather than between words. As a result, compound words like "học sinh" (student) are split into their component syllables "học" and "sinh" rather than being kept as a single token. This is a known limitation of whitespace-based word splitting for Vietnamese.

Per-shard word frequencies. Word and n-gram frequencies are computed per source shard, not aggregated globally. To get language-level frequencies, aggregate with sum(frequency) GROUP BY word in DuckDB or any query engine that can read Parquet.

No additional PII filtering. This dataset does not apply any personally identifiable information filtering beyond what was already done upstream by the FineWeb-2 team. Web text inherently contains names, email addresses, and other personal information.

License

ODC-By 1.0 (Open Data Commons Attribution License), following FineWeb-2's license.

Author

Created by Duc-Tam Nguyen (tamnd) as part of the open-index project.

Citation

@misc{fineweb2nlp2026,
  title   = {FineWeb-2 NLP: Sentences, Paragraphs, Words, and N-grams},
  author  = {Nguyen, Duc-Tam},
  year    = {2026},
  url     = {https://huggingface.co/datasets/open-index/fineweb-2-nlp},
  note    = {Derived from FineWeb-2 (HuggingFaceFW/fineweb-2)}
}

@article{penedo2025fineweb2,
  title   = {FineWeb2: One Pipeline to Scale Them All},
  author  = {Guilherme Penedo and others},
  year    = {2025},
  eprint  = {2506.20920},
  archivePrefix = {arXiv}
}

---

Last updated: 2026-04-15 03:59 UTC