sozkz-corpus-raw-kk-multi-v1

Kazakh text corpus collected from 7 public sources: web crawls, parallel corpora, reviews, and benchmarks. No deduplication applied — raw cleaned texts.

Stats

Field	Value
Total texts	13,149,875
Format	Parquet (19 shards)
Columns	text, source
Cleaning	NFC normalization, control chars, min 20 chars
Date	2026-02-13
Version	v1
License	Apache 2.0

Sources

Source	HF Dataset	Texts	Description
kazparc_sync	issai/kazparc (sync)	9,632,030	KazParC synthetic parallel corpus, Kazakh side
cc100	statmt/cc100 (kk)	1,721,481	CC-100 CommonCrawl Kazakh
kazparc	issai/kazparc (kazparc)	1,647,560	KazParC natural parallel corpus, Kazakh side
kazsandra	issai/kazsandra	146,253	Kazakh sentiment reviews
wikiann	unimelb-nlp/wikiann (kk)	950	WikiANN NER sentences
belebele	facebook/belebele (kaz_Cyrl)	900	Reading comprehension passages
sib200	Davlan/sib200 (kaz_Cyrl)	701	Topic classification sentences

Usage

from datasets import load_dataset

ds = load_dataset("stukenov/sozkz-corpus-raw-kk-multi-v1", split="train")
print(len(ds))  # 13,149,875
print(ds[0]["source"])  # e.g. "kazparc_sync"

Related repos

• Web corpus (deduplicated): stukenov/sozkz-corpus-dedup-kk-web-v1
• Tokenizer: stukenov/kazakh-bpe-32k

Full collection

Together with other corpora:

Dataset	Texts
kz-transformers/multidomain-kazakh-dataset	12,441,904
sozkz-corpus-dedup-kk-web-v1	9,475,089
sozkz-corpus-raw-kk-multi-v1	13,149,875
Combined	~35M