Dataset Viewer
Auto-converted to Parquet Duplicate
rw
stringlengths
3
73
en
stringlengths
1
34
nandika
I write
wandika
you write
yandika
he or she writes
twandika
we write
nanditse
I wrote
nzandika
I will write
andika
Write!
Ndashaka andika
I want to write
Nshobora andika
I can write
nasoma
I read
wasoma
you read
yasoma
he or she reads
twasoma
we read
nasomye
I have read
nzasoma
I will read
soma
Read!
Ndashaka soma
I want to read
Nshobora soma
I can read
navuga
I speak
wavuga
you speak
yavuga
he or she speaks
twavuga
we speak
navuze
I spoke
nzavuga
I will speak
vuga
Speak!
Ndashaka vuga
I want to speak
Nshobora vuga
I can speak
nateka
I cook
wateka
you cook
yateka
he or she cooks
twateka
we cook
natekaye
I cooked
nzateka
I will cook
teka
Cook!
Ndashaka teka
I want to cook
Nshobora teka
I can cook
nafasha
I help
wafasha
you help
yafasha
he or she helps
twafasha
we help
nafashije
I helped
nzafasha
I will help
fasha
Help!
Ndashaka fasha
I want to help
Nshobora fasha
I can help
nabona
I see
wabona
you see
yabona
he or she sees
twabona
we see
nabonye
I saw
nzabona
I will see
bona
See!
Ndashaka bona
I want to see
Nshobora bona
I can see
nakora
I work
wakora
you work
yakora
he or she works
twakora
we work
nakoze
I worked
nzakora
I will work
kora
Work!
Ndashaka kora
I want to work
Nshobora kora
I can work
ngenda
I go
ugenda
you go
agenda
he or she goes
tugenda
we go
nageze
I went
nzagenda
I will go
genda
Go!
Ndashaka genda
I want to go
Nshobora genda
I can go
nza
I come
uza
you come
aza
he or she comes
tuza
we come
naje
I came
nzaza
I will come
ngwino
Come!
Ndashaka za
I want to come
Nshobora za
I can come
ndya
I eat
urya
you eat
arya
he or she eats
turya
we eat
nariye
I ate
nzarya
I will eat
rya
Eat!
Ndashaka rya
I want to eat
Nshobora rya
I can eat
nanywa
I drink
wanywa
you drink
yanywa
he or she drinks
twanywa
we drink
nanyoye
I drank
nzanywa
I will drink
nywa
Drink!
Ndashaka nywa
I want to drink
Nshobora nywa
I can drink
nashyira
I put
End of preview. Expand in Data Studio

Ice Block curated packs

Nova Opus by Mateza. Authors: Mateza Kayondo and Promesse Rukundo.

Small curated English↔Kinyarwanda JSONL packs used to thicken Ice Block basics (primer / greetings / drills) for v5.5 — not the full ~1.28M lake dump.

Why these packs

Lake audit: ~1.28M pairs were mostly general/legal; primer+greetings were ~0.16%. Later NLLB-heavy continues lost to champion v5.2 (capability en-rw 21/22). Prefer curated basics over more mined mass. After v5.5 curated ingest (live): primer 4,582, greetings 162.

Sample pairs

en rw
Hello Muraho
Come here Ngwino hano
Sit down Icara hasi

Files

Path Role
curated/greetings/pack.jsonl Greetings
curated/primer/everyday.jsonl Everyday primer
curated/conjugation/paradigms.jsonl Verb paradigms
curated/math/drills.jsonl Number / math drills
curated/questions/drills.jsonl Question drills
curated/du-everyday/short_primer.jsonl Short DU-style primer lines
curated/du-everyday/short_greetings.jsonl Short greetings (may be sparse)
hf download mateza/ice-block-curated-packs --repo-type dataset \
  --include 'curated/greetings/pack.jsonl' --local-dir ./hf-curated
ls data/curated   # in an Ice-Block clone

Citation

@misc{mateza-opus-ft-v3,
  title        = {Mateza OPUS-FT v3: English-Kinyarwanda translation},
  author       = {Kayondo, Mateza and Rukundo, Promesse},
  year         = {2026},
  note         = {Fine-tuned from Helsinki-NLP OPUS-MT en-rw and rw-en. Organization: Mateza.},
  url          = {https://huggingface.co/mateza/mateza-opus-ft-v3},
  howpublished = {Hugging Face}
}

Not included

The full Ice Block Postgres dump / raw NLLB TSVs are not published here.

Downloads last month
34