GitHub yakavhura budiriro mukushandiswa kwemichina yekudzidza yekutsvaga kodhi uye kuongorora

GitHub kuunzwa ronga CodeSearchNet, iyo yakagadzirira michina yekudzidza modhi uye seti yedata inodiwa pakupatsanura, kurongedza uye kuongorora kodhi mumitauro yakasiyana-siyana yekuronga. CodeSearchNet, yakafanana ne IMAGEnet, inosanganisira muunganidzwa wakakura wemakodhi snippets ane zvirevo zvinogadzira izvo kodhi inoita. Zvikamu zvekudzidzisa modhi uye mienzaniso yekushandisa CodeSearchNet yakanyorwa muPython uchishandisa Tensorflow chimiro uye inoparadzirwa ne pasi peMIT rezinesi.

Pakuumba CodeSearchNet, matekinoroji ekunyora emutauro wechisikigo akashandiswa, zvichiita kuti masisitimu ekudzidza emuchina atore pfungwa kwete chete maumbirwo emhando, asiwo zvinoreva zviito zvinoitwa nekodhi. Iyo GitHub system kuiswa muzviedzo pakuronga semantic kodhi yekutsvaga uchishandisa mibvunzo pa mutauro wechisikigo (somuenzaniso, kana uchikumbira "kuronga runyoro rwetambo", kodhi nekushandiswa kweiyo algorithms inoenderana inoratidzwa).

Iyo dataset yakatsanangurwa inosanganisira inodarika mamirioni maviri kodhi-yekutaura zvinongedzo, zvakagadzirirwa zvichibva pane zvakabva zvinyorwa zvemaraibhurari akavhurika aripo. Iyo kodhi inovhara yakazara sosi mameseji ega ega mabasa kana nzira, uye chirevo chinotsanangura zviito zvinoitwa nebasa (yakadzama zvinyorwa zvinopihwa). Parizvino, datasets dzakagadzirirwa Python, JavaScript, Ruby, Go, Java uye PHP. Mienzaniso inopihwa yekushandisa yakarongwa dhataseti yekudzidzisa akasiyana marudzi eneural network, kusanganisira Neural-Bag-Ye-Mashoko, RNN, Kuzvitarisa (BERT) uye 1D-CNN+Self-Attention Hybrid.

Kugadzira nzira dzekutsvaga dzemitauro, seti yeCodeSearchNet Dambudziko yakagadziridzwa, kusanganisira
99 zvakajairika mibvunzo ine zviuru zvina zvinotsanangurwa nenyanzvi zvinotsanangura zvingango sungirwa kodhi muCodeSearchNet Corpus dataset, inovhara nzira nemaitiro emamiriyoni matanhatu (set size anenge 20 GB). Iyo CodeSearchNet Dambudziko rinogona kushanda sebhenji rekuongorora kushanda kwedzimwe nzira dzekutsvaga kodhi yemutauro wechisikigo. Kushandisa zvishandiso KubeFlow gadzirira
muenzaniso kodhi yekutsvaga injini.

Source: opennet.ru

Voeg