Kalaallisut (kl) 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 sizeklwiki sample
original thule kulturi ukiut 0000 kingornatigut inuit avanersuaq tikeqqaarpaa, inuit-lu issittumi ullumikkut inuusut siuleraat. qaanaaq inuusoqaaq oqaluttuaris
singles * 0000 every little part of me * 0000 shout (out love will be the light) * 0000 completely fallen * 0000 it's a wonderful feeling
naalagaaffiit peqatigiit inuisa isumaqatigiissutikkut uppernarsaqqimmassuk inuttut pisinnaatitaaffiit tunngaviusut, inuup ataqqinassusiata erianassusi
1000 ▁t h u le ▁k ul tu ri ▁uki ut ▁0000 ▁king orn atigut ▁inuit ▁av an ersu aq ▁t ik eqq aar paa , ▁inuit - lu ▁i ss ittu mi ▁ullumi kkut ▁inu usut ▁si uler aat . ▁q aan aaq ▁inu usoq aaq ▁oqaluttu aris
▁s ing le s ▁* ▁0000 ▁e ver y ▁l itt le ▁p art ▁of ▁m e ▁* ▁0000 ▁s ho ut ▁( o ut ▁l ov e ▁w ill ▁b e ▁the ▁ lig ht ) ▁* ▁0000 ▁c om p le t el y ▁f all en ▁* ▁0000 ▁it ' s ▁a ▁w on der f ul ▁f e el ing
▁naalag aaffi it ▁p eqatigi it ▁inu is a ▁isumaq atigi iss ut i kkut ▁upp ern ars aqq i mma ssu k ▁in uttut ▁pisinnaatit aaffi it ▁tunngav i usut , ▁inuup ▁at aqq in ass usi ata ▁ eri an ass usi
3000 ▁th ule ▁kultu ri ▁ukiut ▁0000 ▁kingorn atigut ▁inuit ▁av an ersuaq ▁tik eqqaar paa , ▁inuit - lu ▁iss ittumi ▁ullumikkut ▁inu usut ▁si uler aat . ▁qaan aaq ▁inu usoq aaq ▁oqaluttuaris
▁sing les ▁* ▁0000 ▁e ver y ▁l itt le ▁part ▁of ▁me ▁* ▁0000 ▁s ho ut ▁( o ut ▁l ov e ▁w ill ▁be ▁the ▁ lig ht ) ▁* ▁0000 ▁com ple t el y ▁f all en ▁* ▁0000 ▁it ' s ▁a ▁w on der f ul ▁fe el ing
▁naalagaaffiit ▁peqatigi it ▁inu isa ▁isumaqatigi iss uti kkut ▁uppernars aqq i mma ssuk ▁inuttut ▁pisinnaatit aaffiit ▁tunngavi usut , ▁inuup ▁ataqq in ass usi ata ▁ eri an ass usi
5000 ▁thule ▁kulturi ▁ukiut ▁0000 ▁kingorn atigut ▁inuit ▁av an ersuaq ▁tik eqqaar paa , ▁inuit - lu ▁iss ittumi ▁ullumikkut ▁inu usut ▁si uler aat . ▁qaanaaq ▁inu usoq aaq ▁oqaluttuaris
▁sing les ▁* ▁0000 ▁e ver y ▁l itt le ▁part ▁of ▁me ▁* ▁0000 ▁sho ut ▁( out ▁love ▁will ▁be ▁the ▁ lig ht ) ▁* ▁0000 ▁com ple tel y ▁f allen ▁* ▁0000 ▁it ' s ▁a ▁w on der f ul ▁fe el ing
▁naalagaaffiit ▁peqatigiit ▁inu isa ▁isumaqatigi iss uti kkut ▁uppernars aqq i mma ssuk ▁inuttut ▁pisinnaatit aaffiit ▁tunngavi usut , ▁inuup ▁ataqqin assusi ata ▁eri an assusi