Tahitian (ty) subword embeddings

Vocab size vocab model 25 dim 50 dim 100 dim 200 dim 300 dim
1000 vocab model txt | bin
bokeh | umap | matrix
txt | bin
bokeh | umap | matrix
txt | bin
bokeh | umap | matrix
txt | bin
bokeh | umap | matrix
txt | bin
bokeh | umap | matrix
3000 vocab model txt | bin
bokeh | umap | matrix
txt | bin
bokeh | umap | matrix
txt | bin
bokeh | umap | matrix
txt | bin
bokeh | umap | matrix
txt | bin
bokeh | umap | matrix
5000 vocab model txt | bin
bokeh | umap | matrix
txt | bin
bokeh | umap | matrix
txt | bin
bokeh | umap | matrix
txt | bin
bokeh | umap | matrix
txt | bin
bokeh | umap | matrix

Embedding matrix plots

Training corpus sample, encoded with different BPE vocabulary sizes

Vocab sizetywiki sample
original aita e nanumiti no te âvaè, no te mahana rā. no reira, e pāhee noa ā te tai i te poìpoì roa e i te ahiahi, e pānunu te tai i tuìraa pō e i te avatea.
goldingare láse zse nyásze, nyovergij strommij lá dzsávre di császte. lipo-rimnigueij kinyáresz de vácse, áles párlámingszej jilyngász lá prizenyijpov
i ’europa te fenua heremani (, ) . ’o berlin te ’oire pū. ’o joachim gauck te peretiteni, te reo heremani to na reo.
1000 ▁aita ▁e ▁na nu m iti ▁no ▁te ▁âva è , ▁no ▁te ▁mahana ▁rā . ▁no ▁reira , ▁e ▁pā he e ▁noa ▁ā ▁te ▁tai ▁i ▁te ▁po ì po ì ▁roa ▁e ▁i ▁te ▁a hia hi , ▁e ▁pā nu nu ▁te ▁tai ▁i ▁tu ì raa ▁pō ▁e ▁i ▁te ▁a va tea .
▁g ol d ing a re ▁lá se ▁z se ▁ny ász e , ▁ny o ver gi j ▁st ro m mi j ▁lá ▁d z sá vre ▁di ▁c s ász te . ▁li po - ri m ni g u ei j ▁k iny áresz ▁de ▁v á c se , ▁á les ▁p á rl á m ing sz e j ▁j il y ng ász ▁lá ▁p ri z eny i j po v
▁i ▁’ europa ▁te ▁fenua ▁heremani ▁(, ▁) ▁. ▁’ o ▁b er lin ▁te ▁’ oire ▁pū . ▁’ o ▁j oa c hi m ▁g au ck ▁te ▁peretiteni , ▁te ▁reo ▁heremani ▁to ▁na ▁reo .
3000 ▁aita ▁e ▁na numiti ▁no ▁te ▁âvaè , ▁no ▁te ▁mahana ▁rā . ▁no ▁reira , ▁e ▁pā he e ▁noa ▁ā ▁te ▁tai ▁i ▁te ▁po ì po ì ▁roa ▁e ▁i ▁te ▁a hia hi , ▁e ▁pā nunu ▁te ▁tai ▁i ▁tu ìraa ▁pō ▁e ▁i ▁te ▁ava tea .
▁gold ingare ▁láse ▁zse ▁ny ásze , ▁ny over gij ▁st rommij ▁lá ▁dzsá vre ▁di ▁cs ászte . ▁lipo - ri mni gueij ▁kiny áresz ▁de ▁v ácse , ▁áles ▁p árlám ing szej ▁jily ngász ▁lá ▁priz enyij po v
▁i ▁’ europa ▁te ▁fenua ▁heremani ▁(, ▁) ▁. ▁’ o ▁b erlin ▁te ▁’ oire ▁pū . ▁’ o ▁j oachim ▁g auck ▁te ▁peretiteni , ▁te ▁reo ▁heremani ▁to ▁na ▁reo .
5000 ▁aita ▁e ▁nanumiti ▁no ▁te ▁âvaè , ▁no ▁te ▁mahana ▁rā . ▁no ▁reira , ▁e ▁pāhee ▁noa ▁ā ▁te ▁tai ▁i ▁te ▁poìpoì ▁roa ▁e ▁i ▁te ▁ahiahi , ▁e ▁pānunu ▁te ▁tai ▁i ▁tuìraa ▁pō ▁e ▁i ▁te ▁avatea .
▁goldingare ▁láse ▁zse ▁nyásze , ▁nyovergij ▁strommij ▁lá ▁dzsávre ▁di ▁császte . ▁lipo - rimnigueij ▁kinyáresz ▁de ▁vácse , ▁áles ▁párlámingszej ▁jilyngász ▁lá ▁priz enyij po v
▁i ▁’ europa ▁te ▁fenua ▁heremani ▁(, ▁) ▁. ▁’ o ▁berlin ▁te ▁’ oire ▁pū . ▁’ o ▁joachim ▁gauck ▁te ▁peretiteni , ▁te ▁reo ▁heremani ▁to ▁na ▁reo .