Dataset Viewer
Auto-converted to Parquet Duplicate
latin
stringlengths
1
154
khmer
stringlengths
1
116
persona
stringclasses
5 values
source
stringclasses
6 values
ruoch preah vihea gesang sala sang salarien i jeung
រួច ព្រះវិហារ គេសង់សាលា សង់សាលារៀនអីចឹង
lazy
hplt
tou
ទៅ
careful
hplt
preahong kpg cham
ព្រះអង្គ កំពង់ចាម
mixed
hplt
lok tov sampot pratan kat ksebo eichoeng tov pontae knong
លោកទៅសម្ពោធប្រទានកាត់ខ្សែបូអីចឹងទៅប៉ុន្តែក្នុង
careful
hplt
banda night
បណ្តា night
careful
hplt
Monos tang'oh dael mork juabjum neng ter
មនុស្សទាំងអស់ដែលមកជួបជុំហ្នឹង តើ
mixed
hplt
Del tverka
ដែលធ្វើការ
lazy
hplt
ng nak na dl cheunh prakkahchraen cheang
ហ្នឹង អ្នកណាៗដែលចេញប្រាក់កាសច្រើនជាង
mixed
hplt
ge neak na tel oeeee
គេអ្នកណាដែល អើ
lazy
hplt
!
!
mixed
hplt
youkjoettokdak jang ke nhoeng ke niyeai thaa touch
យកចិត្តទុកដាក់ជាងគេហ្នឹង គេនិយាយថា ដូច
mixed
hplt
som'dat loukthaa somdat uv mean brehtey rik ray
សម្តេចលោកថា សម្តេចឪមានព្រះទ័យ រីករាយ
lazy
hplt
treika nah neung nurse os konchau del ban ka sang salariennn
ត្រេកអរណាស់ និង nurse អស់កូនចៅ ដែលបានកសាងសាលារៀន
mixed
hplt
thoumdom nis l'or sat thcomthcaeung nih som ogoun
ធំដុំនេះ ល្អស្អាត ថ្កុំថ្កើងនេះ សុំអរគុណ
lazy
hplt
teang os
ទាំងអស់
lazy
hplt
knea
គ្នា
formal
hplt
dl ban
ដែលបាន
mixed
hplt
jouy pout day knea sponsor tver nih
ជួយពូតដៃគ្នា sponsor ធ្វើនេះ
mixed
hplt
cheapises akun chompuors lok chavvai khaet keu
ជាពិសេសអរគុណចំពោះលោកចៅហ្វាយខេត្ត គឺ
mixed
hplt
Lok chaohvay khait nus ke thum cheang ge phderm karneung nung cheapises
លោកចៅហ្វាយខេត្តនោះ គេធំជាងគេផ្តើមការហ្នឹង ហ្នឹងជាពិសេស
careful
hplt
rouch neang kangnha
រួចនាងកញ្ញា
lazy
hplt
prakan tha samrab tae samdach preahpramoukroat samrab samdach kor ban
ប្រកាន់ថា សម្រាប់តែសម្តេចព្រះប្រមុខរដ្ឋ សម្រាប់សម្តេចក៏បាន
careful
hplt
yerng kor prae bandocchea yerng tov aina2 bon tean ey
យើងក៏ប្រើបាន ដូចជាយើងទៅឯណាៗបុណ្យទានអី
careful
hplt
reu yoeng tovjea phnhiav pounleu gei ey yaeng nikyiay thar
ឬយើងទៅជាភ្ញៀវពន្លឺគេអី យើងនិយាយថា
mixed
hplt
knym som orkun bongpoun teangos knaea
ខ្ញុំសុំអរគុណបងប្អូនទាំងអស់គ្នា
lazy
hplt
chea pi sis ker ming mau ming mau
ជាពិសេសគឺ មីងម៉ៅ មីងម៉ៅ
mixed
hplt
kot chad kar nung
គាត់ចាត់ការហ្នឹង
mixed
hplt
Kueu Tha Koat Noeay Hat Cheang
គឺថាគាត់នឿយហត់ជាង
formal
hplt
ke nas naego!
គេណាស់ នែក!
lazy
hplt
Noeng Ka Yeung Niyeay Ban
ហ្នឹងក៏យើងនិយាយបាន
formal
hplt
ker
គឺ
mixed
hplt
tha doylaek ker tha
ថាដោយឡែក គឺថា
careful
hplt
specialiste prohelchea sdab ban haey nang kanhnha.
specialiste ប្រហែលជាស្តាប់បានហើយនាងកញ្ញា។
careful
hplt
ponnung
ប៉ុណ្ណឹង
careful
hplt
haeuy atma som dah sray ponnoeng
ហើយអាត្មាសុំដោះស្រាយប៉ុណ្ណឹង
formal
hplt
haeuy neang kom chngal kom chngal
ហើយ នាងកុំឆ្ងល់ កុំឆ្ងល់
formal
hplt
peses heuy veses noh.
ពិសេសហើយវិសេសនោះ។
mixed
hplt
he!
អ្ហេ៎!
careful
hplt
neang nih niyeay nis atma meun meul
នាងនេះនិយាយនេះអាត្មាមិនមើល
lazy
hplt
te tha wifi tae preaek leap lue
ទេ ថា wifi តែ ព្រែកលៀប ឮ
lazy
hplt
wtyo ta domneung piseh eytha muen suv chong sdab
វិទ្យុថា ដំណឹងពិសេស អីថាមិនសូវចង់ស្តាប់
lazy
hplt
tei khoeng nas mej kor
ទេ ខឹងណាស់ ម៉េចក៏
careful
hplt
prornhabaeychung kom kung gom ey
ប្រញ៉ាប់អីចឹង កុំខឹងកុំអី
careful
hplt
Khom rean size tos
ខំរៀន size ទៅ
careful
hplt
hg chong cham an nisng
ឯងចង់ចំអន់នាង
mixed
hplt
nungbndeuch hey!
ហ្នឹងបន្តិច ហ្អេ!
mixed
hplt
moe tos kronhoet kronhang
មើល៍ទៅក្រញើតក្រញ៉ាំង
careful
hplt
thop te dol mean pteahsombeng tovthop chluoh
ថប់តែដល់មានផ្ទះសម្បែងទៅ ថប់ឈ្លោះ
mixed
hplt
ng car
នឹង car
mixed
hplt
pdei te anh … doujche baep
ប្តីទេអញ… ដូចជាបែប
mixed
hplt
mutmoam nas
មុតមាំណាស់
careful
hplt
somtous choh!
សុំទោសចុះ!
careful
hplt
neangaeng kom keung noeng a tma dosrai ruoc hoey niyay o soec teos
នាងឯងកុំខឹងនឹងអាត្មា ដោះស្រាយរួចហើយ និយាយឲ្យសើចទេអស់
mixed
hplt
ponnoeng hoei
ប៉ុណ្ណឹងហើយ
careful
hplt
tek thla
ទឹកថ្លា
mixed
hplt
atma leu sou vityu somtouh
អាត្មាឮសូរវិទ្យុ សុំទោស
mixed
hplt
dal
ដល់
formal
hplt
Neakan Vityu Pi Thngaiti 14 Khaekakatda Nih Nov
អ្នកអានវិទ្យុពីថ្ងៃទី14 ខែកក្កដា នេះ នៅ
formal
hplt
maong 21 kanla an moen doeng jea trang roeang i te leu Tbong khmum
ម៉ោង21កន្លះ អានមិនដឹងជា ត្រង់រឿងអីទេ ឮ ត្បូងឃ្មុំ
careful
hplt
otdar meanchey wanhjana prorwanhjana DM ppontae ke pa de saet no dak
ឧត្តរមានជ័យ វញ្ចនៈ ប្រវញ្ចនា DM ប៉ុន្តែគេបដិសេធ ន ដាក់
careful
hplt
ao tv keta pravanh kapravanh
ឲ្យទៅ គេថា ប្រវ័ញ្ច ការប្រវ័ញ្ច
mixed
hplt
ruojtovan leu tha [ wanhoa phone
រួចទៅអានឮថា [វ៉ាញ់ phone
mixed
hplt
jak nea ]
ចាក់ នា]
lazy
hplt
wanhjana joh mech lok nak an mech
វញ្ចនា ចុះម៉េចលោកអ្នកអាន ម៉េច
mixed
hplt
bro kor muen keut kor chehte raoel
bro ក៏មិនគិត ក៏ចេះតែរអិល
mixed
hplt
mord mun pi bak sahasocho
មាត់ មិនពិបាកសោះអក្សរ ច
mixed
hplt
nov pileu no ruoc piak dai muy mat beu cha
នៅពីលើ ន រួចពាក្យតែមួយម៉ាត់ បើ ច
careful
hplt
Dak Cheung No Ke An Tha Chna – Chna Choh
ដាក់ជើង ន គេអានថា ច្ន – ច្នា ចុះ
formal
hplt
meuc touche [ vanho
ម៉េចទៅជា [វាញ់
mixed
hplt
jak nea ] tovwoenh
ចាក់ នា] ទៅវិញ
mixed
hplt
kor mun tha [ weanho chak na
ក៏មិនថា [វាញ់ ចាក់ ណា
careful
hplt
]
]
careful
hplt
woangnea prwang nea. inbox
វ័ញ្ចនា ប្រវ័ញ្ចនា។ inbox
careful
hplt
douj mitona COD eijoeng
ដូចមិថុនា COD អីចឹង
careful
hplt
[ mi to na ]
[មិ ថុ ណា]
mixed
hplt
atma ler nor!
អាត្មាឮន៏!
mixed
hplt
karsomrach neko!
ការសម្រេច នែក!
lazy
hplt
samdey samdav samdach sap preah reach harutey nah
សម្តីសំដៅសម្តេចសព្វព្រះរាជហឫទ័យណាស់
careful
hplt
pontaitorh an
ប៉ុន្តែទាស់អាន
lazy
hplt
khe [ me to
ខែ [មិ ថុ
mixed
hplt
ea ] he! Update
នា] អ្ហែ៎! Update
lazy
hplt
loukneng doochchea kmean sdap ke sos ge
លោកហ្នឹងដូចជាគ្មានស្តាប់គេសោះ គេ
lazy
hplt
an [ mi tho na ]
អាន [មិ ថុ ន៉ា]
mixed
hplt
mithona neng meun bach nonabongkop
មិថុនាហ្នឹង មិនបាច់នរណាបង្គាប់
lazy
hplt
ke ta mi to
គេថា មិថុ
careful
hplt
to slow haeuy noeng nor via
ថុ slow ហើយនឹង ន វា
careful
hplt
phjuapknea thor dak jerng PassApp no Kompot thnor
ភ្ជាប់គ្នា ថ ដាក់ជើង PassApp ន កំពត ថ្ន
mixed
hplt
– thna ge
– ថ្នា គេ
careful
hplt
tha dnang
ថា ថ្នង
careful
hplt
ke tha tnang her!
គេថា ថ្នាំង ហ៊ឺ!
lazy
hplt
ke htaa thnam ke mun thaa
គេថា ថ្នាំ គេមិនថា
mixed
hplt
thnong a na chos mech toujea mithona tou vuenh
ធ្នង ឯណា ចុះម៉េចទៅជា មិថុនា ទៅវិញ
mixed
hplt
toech louk tha juh baen na mun ao ge tha na yangmuch
តិចលោកថាចុះបើន នា មិនឲ្យគេថា នា យ៉ាងម៉េច
careful
hplt
?
?
mixed
hplt
ruoch choh mean tae preahput sasnea
រួចចុះមានតែ ព្រះពុទ្ធសាស្ន៊ា
formal
hplt
Teat hay naeko!
ទៀតហើយ នែក!
lazy
hplt
a nis no dae
អានេះ ន ដែរ
mixed
hplt
putsasnea,
ពុទ្ធសាស្ន៊ា,
careful
hplt
crasong khossnea ka, … horcor!
ក្រសួងឃោស្ន៊ាការ, … អ្ហ៎ក!
lazy
hplt
vea'sna neng kor va nor
វាសនា ហ្នឹងក៏វា ន
mixed
hplt
End of preview. Expand in Data Studio

Samdei Type Pretraining Data

These are the exact pretraining (stage-1) pairs of the Samdei Type romanized-Khmer → Khmer-script converters: 19.3M (romanized input, Khmer target) pairs. Every Khmer target is real text: 1.41M sentences from HPLT v2, FineWeb-2, Khmer Wikipedia and ALT, plus 2,483 chat sentences written by Gemma 4. The romanized input is made by a rule-based generator that spells each piece of a sentence as one of four typing personas would type it, and for a small part by Gemma 4.

They trained the pretrained checkpoints samdei/samdei-type-base-pretrained and samdei/samdei-type-mini-pretrained, from which Samdei Type Base and Mini were fine-tuned. The fine-tuning and adaptation data (stages 2–3) is in samdei/samdei-type-finetuning-data.

Paper: technical report. Inference code: github.com/Samdei-Labs/samdei-type. Test set: samdei/samdei-type-bench.

Configs and files

The data is organized by what it is, not by model. recipes.json gives the exact mix of each model (files, repeats, steps, learning rate), read from the training arguments stored in the checkpoints.

Config Files (rows) What it is Used by
generator_v2 (default) train_v2_s3 (6,501,912) web sentences and the chat sentences, spelled by the second generator version Mini
generator_v1 train_web_s1 (6,289,947), train_web_s2 (6,288,484) web sentences, spelled twice by the first generator version (two independent samples) Base; the channel model (train_web_s1)
llm_chat_sentences train_gemma_s1 (189,431) the 2.4k LLM-written chat sentences, spelled by the first generator version, ×40 Base; the channel model
llm_romanized train_gemrom (5,715) Gemma 4 romanizations of 2,000 real sentences (×3 copies with random casing) Mini (listed ×3 in its mix); the channel model
Model Stage-1 mix Pairs
Base (18.2M) train_web_s1 + train_web_s2 + train_gemma_s1 12,767,862
Mini (7.1M) train_v2_s3 + train_gemrom ×3 6,519,057
channel model (Khmer → Latin, used in Base's reranker) train_web_s1 + train_gemma_s1 + train_gemrom 6,485,093

Why two generator versions. Base was pretrained first, with the first generator version. The second version adds:

  • English chat words and place names inserted on both sides;
  • standalone place-name pairs (persona = place);
  • sentence-final punctuation dropped from both sides together.

It came out of the errors of the first models. Mini was pretrained with it, and both models' later stages use it. Both versions are here, so both models' pretraining can be reproduced exactly.

Validation splits are synthetic: held-out sentences spelled by the same generator, monitored during training. Checkpoints were selected on separate evaluation sets, which are not part of this dataset.

generator_v1/  train_web_s1.parquet  train_web_s2.parquet  train_web_s1.dev.parquet
generator_v2/  train_v2_s3.parquet   train_v2_s3.dev.parquet
llm/           train_gemma_s1.parquet  train_gemrom.parquet
recipes.json

Columns

Column Content
latin romanized input, as the model saw it
khmer Khmer-script target, normalized with khnormal; English words, digits and brand names are copied from the input
persona formal, careful, mixed or lazy (the generator's typing personas, from standard spelling to heavy chat shortcuts and typos); place (gazetteer pair); gemrom0–gemrom3 (Gemma 4 with one of four style prompts)
source where the Khmer target comes from: hplt, fineweb2, wikipedia, alt (a piece of a web sentence); gemma (a piece of an LLM-written chat sentence); gazetteer; gemrom_<x> (a whole sentence from <x>, romanized by Gemma 4)

How to load

from datasets import load_dataset
ds = load_dataset("samdei/samdei-type-pretraining-data", "generator_v2", split="train")
print(ds[0])  # {'latin': ..., 'khmer': ..., 'persona': ..., 'source': ...}

To rebuild a mix exactly, read recipes.json and concatenate its files in order, with repeats.

How the pairs were made

  1. Khmer targets.
    • 1.41M web sentences: HPLT v2 802k (3 of 13 shards of khm_Khmr), FineWeb-2 280k, Khmer Wikipedia 306k (20231101 dump) and the Khmer side of ALT 22k.
    • Each sentence is normalized with khnormal, split and deduplicated.
    • Lines with URLs or e-mail-like strings, and lines that are mostly Latin, are dropped.
    • Sentences that overlap an evaluation set are removed before any pair is made.
    • 2,483 chat sentences were written by Gemma 4 12B, run locally, across 50 chat domains and 4 registers.
  2. The persona generator:
    1. It cuts each sentence into pieces of 1–12 words (mostly 5–6) with khmercut.
    2. It pronounces each word with Google's Khmer pronunciation lexicon (62k words, CC BY 4.0); rule-based G2P covers the rest.
    3. It spells the piece in a sampled persona: formal 8%, careful 32%, mixed 40%, lazy 20%. Personas differ in spelling choices (nh / ny / gn, final consonants, vowels) and in how often they use a hand-written chat lexicon (knhom, kyom or nh for ខ្ញុំ).
    4. It adds noise: glued or split words, dropped vowels, k/g and t/d swaps, mistyped keys.
  3. LLM romanizations. Gemma 4 12B romanized real sentences "the way Cambodians type on Messenger", with 4 style prompts. The Khmer side stays the original sentence.

Test separation.

  • Every sentence of every evaluation set, including Samdei Type Bench, is removed from the corpus and banned in the generator.
  • No Samdei Type Bench message equals a training target. The only shared strings are common single words.

Reproducibility. These files are the exact pairs the released models were trained on, so training can be repeated without regenerating them. The data generator and the training code are not public; the released code (github.com/Samdei-Labs/samdei-type) is for inference only.

Limitations

  • Synthetic inputs. No row is real typing; romanizations follow our four personas and one LLM.
  • Pieces, not messages. Inputs are short (median 26 characters, at most 132).
  • Web noise. About 1.5% of HPLT lines are misordered text extracted from PDFs.
  • Personal data. Web text may contain names, phone numbers or other personal data of the original authors. Lines with URLs or e-mail-like strings were dropped, but no other personal-data filter was run. To have a row removed, open a discussion in the Community tab.
  • LLM text. The Gemma-written chat sentences and romanizations can contain mistakes.

License and attribution

CC BY-SA 4.0 (license text). The rows contain pieces of the source sentences, so the source licenses travel with them: Khmer Wikipedia's share-alike terms set the license of the whole set, and rows from the other sources also stay under their own terms below.

Source Side License
HPLT v2 (HPLT/HPLT2.0_cleaned, khm_Khmr) Khmer CC0 packaging; Common Crawl terms of use apply to the text
FineWeb-2 (HuggingFaceFW/fineweb-2, khm_Khmr) Khmer ODC-By 1.0; Common Crawl terms of use apply to the text
Khmer Wikipedia (wikimedia/wikipedia, 20231101.km) Khmer CC BY-SA 4.0 (share-alike)
Asian Language Treebank (ALT), Khmer side Khmer CC BY 4.0
LLM-written chat sentences (ours, Gemma 4) Khmer CC BY 4.0 (Gemma 4: Apache-2.0)
Gemma 4 romanizations (ours) Latin Apache-2.0 model output; Khmer side as its source
Google Khmer pronunciation lexicon Latin (generator) CC BY 4.0
Chat lexicon, place names, English chat words (ours) Latin (generator) CC0

Rows whose source is wikipedia or gemrom_wikipedia carry Wikipedia's share-alike terms. They are about 19% of train_web_s1, train_web_s2 and train_v2_s3.

Changes to the source text. The sentences were normalized with khnormal, filtered and deduplicated; the generator cut them into pieces of 1–12 words. Each Khmer target is paired with a romanization in the latin column (see "How the pairs were made").

Citation

@misc{samdei-type-2026,
  title  = {Samdei Type: Separating What Is Written from How It Is Spelled for Romanized Khmer Transliteration},
  author = {Mengsay Loem},
  year   = {2026},
  note   = {Technical report},
  url    = {https://github.com/Samdei-Labs/samdei-type/blob/main/paper/samdei-type.pdf}
}

Contact: open a discussion in this repository's Community tab.

Downloads last month
53

Models trained or fine-tuned on samdei/samdei-type-pretraining-data

Collection including samdei/samdei-type-pretraining-data