Dataset Viewer
Auto-converted to Parquet Duplicate
latin
stringlengths
1
132
khmer
stringlengths
1
105
persona
stringclasses
5 values
source
stringclasses
6 values
>>Quick Selection Tools: prae
>>Quick Selection Tools: αž”αŸ’αžšαžΎ
mixed
hplt
SOMRAP TWER KA SELECT TOU LER
αžŸαŸ†αžšαžΆαž”αŸ‹αž’αŸ’αžœαžΎαž€αžΆαžš Select αž‘αŸ…αž›αžΎ
careful
hplt
rouppab dael man laknahjbah vea manlaknah
αžšαžΌαž”αž—αžΆαž– αžŠαŸ‚αž›αž˜αžΆαž“ αž›αž€αŸ’αžαžŽαŸ‡ αž…αŸ’αž”αžΆαžŸαŸ‹ αžœαžΆαž˜αžΆαž“ αž›αž€αŸ’αžαžŽαŸ‡
lazy
hplt
ngeai sruol nhing japrorhosaaaa
αž„αžΆαž™ αžŸαŸ’αžšαž½αž› αž“αž·αž„ αž†αžΆαž”αŸ‹αžšαž αžŸαŸ
mixed
hplt
daoy louk nak gruan dai choch mau
αžŠαŸ„αž™ αž›αŸ„αž€ αž’αŸ’αž“αž€ αž‚αŸ’αžšαžΆαž“αŸ‹ αžαŸ‚ αž…αž»αž…αž˜αŸ‰αŸ…
lazy
hplt
khang jveng jap new tianh tow
αžαžΆαž„αž†αŸ’αžœαŸαž„ αž…αžΆαž”αŸ‹ new αž‘αžΆαž‰ αž‘αŸ…
mixed
hplt
LOE TOMRONGROUPPEAB LOK NEAK
αž›αžΎ αž‘αŸ†αžšαž„αŸ‹ αžšαžΌαž”αž—αžΆαž– αž›αŸ„αž€ αž’αŸ’αž“αž€
lazy
hplt
bnnrh ( travte thveu bonti mdang2 kom ao huorh
αž”αŸ‰αž»αž“αŸ’αž“αŸ„αŸ‡(αžαŸ’αžšαžœαžαŸ‚αž’αŸ’αžœαžΎαž”αž“αŸ’αž‘αž·αž˜αŸ’αžαž„αŸ—αž€αž»αŸ†αž²αŸ’αž™αž αž½αžŸ
lazy
hplt
kangkrav roppeab )
αžαžΆαž„αž€αŸ’αžšαŸ…αžšαžΌαž”αž—αžΆαž–)
lazy
hplt
>> Magic
>> Magic
mixed
hplt
wand Tools prae prasaTools nih
wand Tools αž”αŸ’αžšαžΎ αž”αŸ’αžšαžΆαžŸαŸ‹Tools αž“αŸαŸ‡
mixed
hplt
knong ka Select
αž€αŸ’αž“αž»αž„ αž€αžΆαžš Select
careful
hplt
leurouppheap dael
αž›αžΎ αžšαžΌαž”αž—αžΆαž– αžŠαŸ‚αž›
mixed
hplt
mean mean
αž˜αžΆαž“ αž˜αžΆαž“
mixed
hplt
phtey Background por no tae muy haey men
αž•αŸ’αž‘αŸƒ Background αž–αžšαžŽαŸ αžαŸ‚αž˜αž½αž™ αž αžΎαž™αž˜αžΆαž“
lazy
hplt
laknah cbas vea
αž›αž€αŸ’αžαžŽαŸ‡ αž…αŸ’αž”αžΆαžŸαŸ‹ វអ
careful
hplt
mian laknah ngii ng jabror ho so bomput douy
αž˜αžΆαž“ αž›αž€αŸ’αžαžŽαŸ‡ αž„αžΆαž™ αž“αž·αž„ αž†αžΆαž”αŸ‹ រហស័ αž”αŸ†αž•αž»αž αžŠαŸ„αž™
lazy
hplt
louk neakgroan'tae choch tov loe rouppap
αž›αŸ„αž€ αž’αŸ’αž“αž€ αž‚αŸ’αžšαžΆαž“αŸ‹αžαŸ‚ αž…αž»αž… αž‘αŸ…αž›αžΎ αžšαžΌαž”αž—αžΆαž–
careful
hplt
brl nuhwa ng choen
αž–αŸαž›αž“αŸ„αŸ‡ វអ αž“αž·αž„ αž…αŸαž‰
mixed
hplt
Select Vaccine douy kluon tov ler
Select Vaccine αžŠαŸ„αž™ αžαŸ’αž›αž½αž“ αž‘αŸ…αž›αžΎ
careful
hplt
roupiab teanglay na dail
αžšαžΌαž—αžΆαž– αž‘αžΆαŸ†αž„αž‘αžΆαž™ ណអ αžŠαŸ‚αž›
mixed
hplt
mean pon sradeang kna
αž˜αžΆαž“ αž–αžŽαŸ αžŸαŸ’αžšαžŠαŸ€αž„αž‚αŸ’αž“αžΆ
mixed
hplt
novknong
αž“αŸ…αž€αŸ’αž“αž»αž„
mixed
hplt
krom Rectangular Marquee Tools: prae
αž€αŸ’αžšαž»αž˜ Rectangular Marquee Tools:αž”αŸ’αžšαžΎ
mixed
hplt
somrab tvoe kar katrouppheap mean reang
αžŸαŸ†αžšαžΆαž”αŸ‹ αž’αŸ’αžœαžΎαž€αžΆαžš αž€αžΆαžαŸ‹ αžšαžΌαž”αž—αžΆαž– αž˜αžΆαž“ αžšαžΆαž„
mixed
hplt
buon jrong nhing mean booking ringjea
αž”αž½αž“ αž‡αŸ’αžšαž»αž„ αž“αž·αž„ αž˜αžΆαž“ booking αžšαžΆαž„ αž‡αžΆ
lazy
hplt
Nurse rungvung via mian Tools
nurse αžšαž„αŸ’αžœαž„αŸ‹ αžœαžΆαž˜αžΆαž“ Tools
careful
hplt
cham'nuon 4 dl min leak brand
αž…αŸ†αž“αž½αž“4 αžŠαŸ‚αž› αž˜αžΆαž“ αž›αž€αŸ’αž brand
mixed
hplt
srodiang kna
αžŸαŸ’αžšαžŠαŸ€αž„ αž‚αŸ’αž“αžΆ
lazy
hplt
douc khangkraomea
αžŠαžΌαž… αžαžΆαž„αž€αŸ’αžšαŸ„αž˜αŸˆ
mixed
hplt
>> Single row
>> Single row
careful
hplt
Marquee Tools: Praeu Samrab Kat Rubpheab Chea
Marquee Tools: αž”αŸ’αžšαžΎ αžŸαŸ†αžšαžΆαž”αŸ‹ αž€αžΆαžαŸ‹ αžšαžΌαž”αž—αžΆαž– αž‡αžΆ
formal
hplt
reang trou weng del mean teung
αžšαžΆαž„ αž‘αŸ’αžšαžœαŸ‚αž„ αžŠαŸ‚αž› αž˜αžΆαž“ αž‘αžΉαž„
careful
hplt
jamnuon 1 pech sail (Pixel) ge praeu
αž…αŸ†αž“αž½αž“ 1 αž–αž·αž…αžŸαŸ‚αž›(Pixel)αž‚αŸ αž”αŸ’αžšαžΎ
careful
hplt
vea
វអ
mixed
hplt
somrab sop dveuja sday jnot jnot map
αžŸαŸ†αžšαžΆαž”αŸ‹ αžŸαž»αž” αž’αŸ’αžœαžΎαž‡αžΆ αžŸαŸ’αžαžΆαž™ αž†αŸ’αž“αžΌαžαŸ— map
lazy
hplt
bandouy pi
αž”αžŽαŸ’αžαŸ„αž™αž–αžΈ
mixed
hplt
jveng tea tov sdam
αž†αŸ’αžœαŸαž„ tea αž‘αŸ…αžŸαŸ’αžαžΆαŸ†
careful
hplt
>> Single column Marquee Tools: praeu somrap
>> Single column Marquee Tools: αž”αŸ’αžšαžΎ αžŸαŸ†αžšαžΆαž”αŸ‹
mixed
hplt
kat rouppap jea reang tro
αž€αžΆαžαŸ‹ αžšαžΌαž”αž—αžΆαž– αž‡αžΆαžšαžΆαž„ αž‘αŸ’αžš
mixed
hplt
veng 7am del man toeng
αžœαŸ‚αž„ 7am αžŠαŸ‚αž› αž˜αžΆαž“ αž‘αžΉαž„
careful
hplt
jom nuan 1 pic sael (Pixel) ke proe
αž…αŸ†αž“αž½αž“ 1 αž–αž·αž…αžŸαŸ‚αž›(Pixel)αž‚αŸ αž”αŸ’αžšαžΎ
mixed
hplt
va somrab sob thveujea sdai jnoud
វអ αžŸαŸ†αžšαžΆαž”αŸ‹ αžŸαž»αž” αž’αŸ’αžœαžΎαž‡αžΆ αžŸαŸ’αžαžΆαž™ αž†αŸ’αž“αžΌαž
careful
hplt
Piler joh krom.
αŸ— αž–αžΈαž›αžΎαž…αž»αŸ‡αž€αŸ’αžšαŸ„αž˜αŸ”
careful
hplt
nouknong krom Lasso Tool
αž“αŸ…αž€αŸ’αž“αž»αž„αž€αŸ’αžšαž»αž˜ Lasso Tool
mixed
hplt
nih man
αž“αŸαŸ‡αž˜αžΆαž“
lazy
hplt
Obkornor prohak prohel knea
αž§αž”αž€αžšαžŽαŸ αž”αŸ’αžšαž αžΆαž€αŸ‹ αž”αŸ’αžšαž αŸ‚αž› αž‚αŸ’αž“αžΆ
mixed
hplt
chamnuon weekend bei ke
αž…αŸ†αž“αž½αž“ weekend αž”αžΈ αž‚αŸ
lazy
hplt
Prae wia samrap kat tovtam rouppiap del ke chong
αž”αŸ’αžšαžΎαžœαžΆ αžŸαŸ†αžšαžΆαž”αŸ‹ αž€αžΆαžαŸ‹ αž‘αŸ… តអម αžšαžΌαž”αž—αžΆαž– αžŠαŸ‚αž› αž‚αŸαž…αž„
careful
hplt
baan
αž”αžΆαž“
mixed
hplt
pol gi ge meun jong ban Back groundke teng
αž–αŸ„αž› αž‚αžΈ αž‚αŸαž˜αž·αž“ αž…αž„αŸ‹ αž”αžΆαž“ Back ground αž‚αŸαžαŸ‚αž„
careful
hplt
tae proe Tools torm bi ni.
αžαŸ‚αž”αŸ’αžšαžΎ Tools αž‘αžΆαŸ†αž”αžΈ αž“αŸαŸ‡αŸ”
careful
hplt
Tools teang nuh mean doj
Tools αž‘αžΆαŸ†αž„ αž“αŸ„αŸ‡αž˜αžΆαž“ αžŠαžΌαž…
careful
hplt
ja
αž‡αžΆαŸ–
mixed
hplt
>> Lasso
>> Lasso
careful
hplt
Tool: praeu samrab kat
Tool: αž”αŸ’αžšαžΎαžŸαŸ†αžšαžΆαž”αŸ‹ αž€αžΆαžαŸ‹
formal
hplt
rup phap douy prer mou joch jop
αžšαžΌαž”αž—αžΆαž– αžŠαŸ„αž™ αž”αŸ’αžšαžΎ αž˜αŸ…αž…αž»αž… αž‡αžΆαž”αŸ‹
mixed
hplt
chea nuch mean leakna bat baen
αž‡αžΆ αž“αž·αž…αŸ’αž… αž˜αžΆαž“ αž›αž€αŸ’αžαžŽαŸ‡ αž”αž‘αž”αŸ‚αž“
careful
hplt
TOV TAM JUT COMNOAM
αž‘αŸ…αžαžΆαž˜ αž…αž·αžαŸ’αž αž…αŸ†αž“αžΆαŸ†
mixed
hplt
pel katmun triv leng mau nus tey
αž–αŸαž› αž€αžΆαžαŸ‹ αž˜αž·αž“αžαŸ’αžšαžΌαžœ αž›αŸ‚αž„ αž˜αŸ‰αŸ…αž“αŸ„αŸ‡ αž‘αŸ
mixed
hplt
trov choch mou o choap
αžαŸ’αžšαžΌαžœ αž…αž»αž… αž˜αŸ…αž²αŸ’αž™αž‡αžΆαž”αŸ‹
careful
hplt
cheanech.
αž‡αžΆαž“αž·αž…αŸ’αž…αŸ”
lazy
hplt
>> Polygonal Lasso Tool: Praeu Samrab
>> Polygonal Lasso Tool: αž”αŸ’αžšαžΎαžŸαŸ†αžšαžΆαž”αŸ‹
formal
hplt
Select kat tov leu rouppeab daoy
Select αž€αžΆαžαŸ‹ αž‘αŸ… αž›αžΎ αžšαžΌαž”αž—αžΆαž–αžŠαŸ„αž™
mixed
hplt
preumav choch tou leu
αž”αŸ’αžšαžΎαž˜αŸ‰αŸ… αž…αž»αž… αž‘αŸ… αž›αžΎ
lazy
hplt
vea
វអ
mixed
hplt
mdang mouy domnat2
αž˜αŸ’αžαž„ αž˜αž½αž™ αžαŸ†αžŽαžΆαžαŸ‹αŸ—
careful
hplt
vea mean leaknah chea bantoat roat trang trang
αžœαžΆαž˜αžΆαž“ αž›αž€αŸ’αžαžŽαŸ‡αž‡αžΆ αž”αž“αŸ’αž‘αžΆαžαŸ‹ αžšαžΆαžαŸ‹ αžαŸ’αžšαž„αŸ‹αŸ—
formal
hplt
>> Magnetic Lasso Tool:
>> Magnetic Lasso Tool:
mixed
hplt
proe samrap kat rupphap tov tam
αž”αŸ’αžšαžΎ αžŸαŸ†αžšαžΆαž”αŸ‹ αž€αžΆαžαŸ‹ αžšαžΌαž”αž—αžΆαž– αž‘αŸ… តអម
careful
hplt
tum rung robas vea del yeung changban vea mean peapngeay sruol douy lok
αž‘αŸ†αžšαž„αŸ‹ αžšαž”αžŸαŸ‹αžœαžΆ αžŠαŸ‚αž› αž™αžΎαž„ αž…αž„αŸ‹αž”αžΆαž“ វអ αž˜αžΆαž“ αž—αžΆαž–αž„αžΆαž™ αžŸαŸ’αžšαž½αž› αžŠαŸ„αž™ αž›αŸ„αž€
careful
hplt
neak ticket kruan tae ous tov tam
αž’αŸ’αž“αž€ ticket αž‚αŸ’αžšαžΆαž“αŸ‹ αžαŸ‚ ធូស αž‘αŸ… តអម
mixed
hplt
Tumrung rubhaeuy novpel dol
αž‘αŸ†αžšαž„αŸ‹αžšαžΌαž” αž αžΎαž™ αž“αŸ…αž–αŸαž› αžŠαž›αŸ‹
mixed
hplt
kanleng na man sa peabmeun jbas greanor
αž€αž“αŸ’αž›αŸ‚αž„ αžŽαžΆαž˜αžΆαž“ αžŸαž—αžΆαž– αž˜αž·αž“αž…αŸ’αž”αžΆαžŸαŸ‹ αž‚αŸ’αžšαžΆαž“
mixed
hplt
te chuch
αžαŸ‚αž…αž»αž…
lazy
hplt
mow jweing te suang bnn.
αž˜αŸ…αž†αŸ’αžœαŸαž„ αžαŸ‚ αžŸαž½αž„ αž”αŸ‰αž»αž“αŸ’αž“αŸ„αŸ‡αŸ”
lazy
hplt
>> tek thla Slice Tool: prey
>> αž‘αžΉαž€αžαŸ’αž›αžΆ Slice Tool: αž”αŸ’αžšαžΈ
mixed
hplt
somrab tveu kar kat roubpab
αžŸαŸ†αžšαžΆαž”αŸ‹ αž’αŸ’αžœαžΎαž€αžΆαžš αž€αžΆαžαŸ‹ αžšαžΌαž”αž—αžΆαž–
mixed
hplt
chea
αž‡αžΆ
careful
hplt
pnaik2 deltrov yok hot
αž•αŸ’αž“αŸ‚αž€αŸ— αžŠαŸ‚αž› αžαŸ’αžšαžΌαžœ αž™αž€ hot
careful
hplt
tov prae daembey phchorp
αž‘αŸ… αž”αŸ’αžšαžΎ αžŠαžΎαž˜αŸ’αž”αžΈ αž—αŸ’αž‡αžΆαž”αŸ‹
careful
hplt
tov neungHTML Text nov knung Website
αž‘αŸ… αž“αž·αž„ HTML Text αž“αŸ… αž€αŸ’αž“αž»αž„ Website
careful
hplt
>> prer samrab kat
>> αž”αŸ’αžšαžΎαžŸαŸ†αžšαžΆαž”αŸ‹ αž€αžΆαžαŸ‹
careful
hplt
jam damruem kaem ney rubpheab teangmul
jam αžαŸ†αžšαžΉαž˜ αž‚αŸ‚αž˜ αž“αŸƒ αžšαžΌαž”αž—αžΆαž– αž‘αžΆαŸ†αž„αž˜αžΌαž›
formal
hplt
Posat domrum yok tae reang
αž–αŸ„αž’αž·αŸαžŸαžΆαžαŸ‹ αžαŸ†αžšαžΉαž˜ αž™αž€ αžαŸ‚ αžšαžΆαž„
careful
hplt
bon jrong khang knung tae bon noh robeap brer ngiaysrol lok
αž”αž½αž“αž‡αŸ’αžšαž»αž„αžαžΆαž„ αž€αŸ’αž“αž»αž„αžαŸ‚αž”αŸ‰αž»αž“αŸ’αž“αŸ„αŸ‡ αžšαž”αŸ€αž”αž”αŸ’αžšαžΎ αž„αžΆαž™αžŸαŸ’αžšαž½αž› αž›αŸ„αž€
mixed
hplt
nekkroan te
αž’αŸ’αž“αž€ αž‚αŸ’αžšαžΆαž“αŸ‹αžαŸ‚
mixed
hplt
choch mao chveng oy chop haey
αž…αž»αž… αž˜αŸ‰αŸ…αž†αŸ’αžœαŸαž„ αž’αŸ„αž™αž‡αžΆαž”αŸ‹ αž αžΎαž™
careful
hplt
dianh tow ler ko nleng roubpiab dael louk
αž‘αžΆαž‰ αž‘αŸ…αž›αžΎαž€ αž“αŸ’αž›αŸ‚αž„αžšαžΌαž”αž—αžΆαž– αžŠαŸ‚αž› αž›αŸ„αž€
careful
hplt
nak jang tver ka Crop
αž’αŸ’αž“αž€ αž…αž„αŸ‹ αž’αŸ’αžœαžΎαž€αžΆαžš Crop
careful
hplt
novbontuabmok joch Key
αž“αŸ… αž”αž“αŸ’αž‘αžΆαž”αŸ‹αž˜αž€ αž…αž»αž… Key
careful
hplt
Enter nov pel nus roobpeab krau
Enter αž“αŸ… αž–αŸαž›αž“αŸ„αŸ‡ αžšαžΌαž”αž—αžΆαž– αž€αŸ’αžšαŸ…
careful
hplt
Select raboh size lok neak ningggg
Select αžšαž”αžŸαŸ‹ size αž›αŸ„αž€αž’αŸ’αž“αž€ αž“αž·αž„
mixed
hplt
bat dear novsal te rouppheap nov knung Select ponnua
αž”αžΆαžαŸ‹ dear αž“αŸ…αžŸαž›αŸ‹αžαŸ‚ αžšαžΌαž”αž—αžΆαž– αž“αŸ… αž€αŸ’αž“αž»αž„ Select αž”αŸ‰αž»αž“αŸ’αž“αŸ„αŸ‡
mixed
hplt
. training
αŸ” training
careful
hplt
>> Spot Healing Brush Tool: prer
>> Spot Healing Brush Tool: αž”αŸ’αžšαžΎ
careful
hplt
somrap dvaeka lup bambat cham nuch toch tochdel mean
αžŸαŸ†αžšαžΆαž”αŸ‹ αž’αŸ’αžœαžΎαž€αžΆαžš αž›αž»αž” αž”αŸ†αž”αžΆαžαŸ‹ αž…αŸ†αž“αž»αž… αžαžΌαž…αŸ— αžŠαŸ‚αž› αž˜αžΆαž“
mixed
hplt
NOW LEU ROUBPEAB DOU CHEA MOUN BRCHROUY NG CHAMNUC PSE
αž“αŸ…αž›αžΎ αžšαžΌαž”αž—αžΆαž– αžŠαžΌαž‡αžΆ αž˜αž»αž“ αž”αŸ’αžšαž‡αŸ’αžšαž»αž™ αž“αž·αž„ αž…αŸ†αž“αž»αž… αž•αŸ’αžŸαŸ
mixed
hplt
teat jea jraen teat ….
αŸ— αž‘αŸ€αž αž‡αžΆ αž…αŸ’αžšαžΎαž“ αž‘αŸ€αžβ€¦.
mixed
hplt
>> Patch tool: prer somrab Select
>> Patch tool: αž”αŸ’αžšαžΎαžŸαŸ†αžšαžΆαž”αŸ‹ Select
mixed
hplt
End of preview. Expand in Data Studio

Samdei Type Fine-tuning Data

These are the exact fine-tuning (stage 2) and adaptation (stage 3) pairs of the Samdei Type romanized-Khmer β†’ Khmer-script converters, Samdei Type Mini and Samdei Type Base.

Pretraining (stage 1, samdei/samdei-type-pretraining-data) teaches Khmer words and spelling from 6.5–12.8M generator pairs. These two short stages teach how people actually type in chat:

  • Stage 2: looser LLM spellings, spelling habits mined from public word lists, and chat-like abbreviations.
  • Stage 3: the spoken register and stretched letters.

Each stage is 3–6k steps (10–36 minutes on a laptop).

Every Khmer target is real text, except 2,483 chat sentences written by Gemma 4. Only the romanized input varies, by source.

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

Configs

The data is organized by kind, so each source of spelling noise can be used, inspected or left out on its own. Stages 2 and 3 share one dataset because stage 3 replays the whole stage-2 mix; the stage-3 data is the casual config.

Config Files (rows) What it is Stage Used by
chat_mix (default) ft_v2 (629,878) a chat-heavy sample of generator pairs: more LLM-written chat sentences and ALT, fewer web pieces; English chat words and place names inserted 2, 3 Mini, Base
llm_romanized train_gemrom2 (18,273) Gemma 4 romanizations of 6,382 real sentences, 4 style prompts, Γ—3 copies with random casing 2, 3 (Γ—4 in the mix) Mini, Base
abbreviated ft_v2_abbr3 (139,704), train_gemrom2_abbr3 (16,302) chat-like abbreviated copies of chat_mix and llm_romanized pairs; only the input changes (tov β†’ tv, bong β†’ bg) 2, 3 (Gemma part Γ—2) Mini, Base
mined_tables_chh ft_rules_chh (289,820) new pairs spelled with the table mined from the Chhunneng word list (dictionary-like scheme), alone or with mixed / lazy noise habits 2, 3 Mini
mined_tables_chh_wd ft_rules_chhwd (289,647) made the same way from the same sources, but spelled 70% with the Chhunneng table and 30% with the table mined from Wikidata names (official, name-style scheme) 2, 3 Base
casual ft_casual_noakk (426,946) 21.6k spoken-register web sentences, spelled by the generator with stretched letters (tovvv, baaan) 3 only Mini, Base
synthetic_heldout train_v2_s3.dev (12,232) synthetic held-out pairs monitored during training (validation split) – –

Mini and Base use the same data. The only difference is the mined-table file: Mini uses mined_tables_chh, Base mined_tables_chh_wd.

Mix Files, with repeats Pairs
Stage 2, Base ft_v2 + ft_v2_abbr3 + train_gemrom2 Γ—4 + train_gemrom2_abbr3 Γ—2 + ft_rules_chhwd 1,164,925
Stage 2, Mini the same with ft_rules_chh 1,165,098
Stage 3 the stage-2 mix of the model + ft_casual_noakk 1,591,871 (Base), 1,592,044 (Mini)

recipes.json gives the exact mix of each checkpoint (files in order, repeats, steps, learning rate, initialization), read from the training arguments stored in the checkpoints.

generator/      ft_v2.parquet
llm_romanized/  train_gemrom2.parquet
abbreviated/    ft_v2_abbr3.parquet  train_gemrom2_abbr3.parquet
mined_tables/   ft_rules_chh.parquet  ft_rules_chhwd.parquet
casual/         ft_casual_noakk.parquet
validation/     train_v2_s3.dev.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 how the input was made (below)
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); casual (a piece of a spoken-register HPLT or FineWeb-2 sentence); gazetteer; gemrom_<x> (a whole sentence from <x>, romanized by Gemma 4)
persona Meaning
formal, careful, mixed, lazy the generator's typing personas, from standard spelling to heavy chat shortcuts and typos
place standalone place-name pair from our gazetteer
rules_chh, rules_chh+mixed, rules_chh+lazy Chhunneng-mined table, alone or with the noise habits of mixed / lazy
rules_wd, rules_wd+mixed Wikidata-mined table
…~stretch the persona also stretches letters inside the sentence (stage 3)
gemrom0–gemrom3 Gemma 4 romanization with one of four style prompts
abbr abbreviated copy of another pair

How to load

from datasets import load_dataset
casual = load_dataset("samdei/samdei-type-finetuning-data", "casual", split="train")
abbr = load_dataset("samdei/samdei-type-finetuning-data", "abbreviated", split="train")

To rebuild a mix exactly, read recipes.json and concatenate its files in order, with repeats. Training starts from the stage-1 checkpoints, samdei/samdei-type-base-pretrained and samdei/samdei-type-mini-pretrained.

How the pairs were made

  • chat_mix. The persona generator (second version) on a chat-heavy sample of sentences. The generator works in four steps:
    1. it cuts sentences into pieces of 1–12 words;
    2. it pronounces each word with Google's Khmer pronunciation lexicon;
    3. it spells the piece in a persona (formal 8%, careful 32%, mixed 40%, lazy 20%);
    4. it adds noise (glued or split words, dropped vowels, swaps, typos), English chat words and place names.
  • llm_romanized. Gemma 4 12B, run locally, romanized real sentences "the way Cambodians type on Messenger", 10 sentences per call. The Khmer side stays the original sentence, so the target is always correct, and Gemma's looser spelling acts as human-like noise on the input. 97% of the outputs passed our filters.
  • mined_tables_*. A rule miner learns per-slot spelling tables (onset, vowel, coda, glide, context) by Viterbi alignment and hard EM from public (Khmer word, romanization) lists: the Chhunneng word list (24.9k pairs, MIT) and 1.6k Wikidata person and place names (CC0). Words that are targets of a word-level evaluation set are removed first. The tables are in samdei/samdei-type-resources. The pairs here are new sentences (HPLT, FineWeb-2, ALT, chat sentences) spelled with them.
  • abbreviated.
    • It covers 40% of chat_mix pairs and all llm_romanized pairs.
    • Operations: consonant skeleton (bong β†’ bg), dropped vowels, first + last letter (jong β†’ jg), dropped digraph h (khlang β†’ klang), dropped cluster member.
    • Vowels are dropped only from one-syllable words among the 1,000 most frequent.
    • An abbreviation that the chat lexicon reserves for one word (nh for αžαŸ’αž‰αž»αŸ†) is never produced for another.
  • casual. 21.6k HPLT and FineWeb-2 sentences that contain one of 13 spoken-register words (αž αŸ’αž“αžΉαž„, αž…αžΉαž„, ធី, αžαŸ„αŸ‡, …) as a token; misordered PDF text is dropped. They are spelled with stretched letters by the generator (formal 5%, careful 25%, mixed 35%, lazy 15%, Chhunneng table with mixed habits 20%). No Akkhara-derived spelling rules are used.

Test separation.

  • Every sentence of every evaluation set, including Samdei Type Bench and its chat context, is banned in the generator.
  • The rule miner never sees a test row or a word-level test target.
  • 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. Chat habits are covered only as far as our personas, two mined tables, one LLM and our abbreviation rules reach.
  • Scheme bias of the mined tables. The Chhunneng table follows a dictionary-like scheme and the Wikidata table an official name scheme; neither is chat spelling as such.
  • Pieces, not messages. Inputs are short (median 26 characters).
  • 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
Chhunneng romanized Khmer dataset (Β© 2025 Chrea Chanchhunneng) Latin (mined table) MIT
Wikidata names Latin (mined table) CC0
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 occur only in chat_mix, llm_romanized and abbreviated; the mined_tables_* and casual configs contain none.

The MIT notice of the Chhunneng dataset, which the mined_tables_* tables derive from, is included as LICENSE-chhunneng.txt.

Changes to the source text. The sentences were normalized with khnormal, filtered and deduplicated; the generator cut them into pieces of 1–12 words (the llm_romanized rows keep whole sentences). 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
71

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

Collection including samdei/samdei-type-finetuning-data