{"id":114272,"date":"2024-03-12T18:25:42","date_gmt":"2024-03-12T16:25:42","guid":{"rendered":"https:\/\/prohoster.info\/blog\/novosti-interneta\/proekt-openai-otkryl-transformer-debugger-otladchik-dlya-modelej-mashinnogo-obucheniya"},"modified":"2024-03-12T18:25:42","modified_gmt":"2024-03-12T16:25:42","slug":"proekt-openai-otkryl-transformer-debugger-otladchik-dlya-modelej-mashinnogo-obucheniya","status":"publish","type":"post","link":"https:\/\/prohoster.info\/sq\/blog\/novosti-interneta\/proekt-openai-otkryl-transformer-debugger-otladchik-dlya-modelej-mashinnogo-obucheniya","title":{"rendered":"Projekti OpenAI ka hapur Debugger-in e Transformer-it, nj\u00eb mjet p\u00ebr debug-imin e modeleve t\u00eb m\u00ebsimit t\u00eb makinave","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"
Projekti OpenAI, i angazhuar n\u00eb zhvillimin e projekteve t\u00eb hapura n\u00eb fush\u00ebn e inteligjenc\u00ebs artificiale, publikoi Debugger-in e Transformer-it, i cili \u00ebsht\u00eb i destinuar p\u00ebr analizimin e aktivizimit t\u00eb strukturave n\u00eb modelet gjuh\u00ebsore t\u00eb m\u00ebsimit t\u00eb makinave gjat\u00eb p\u00ebrpunimit t\u00eb t\u00eb dh\u00ebnave t\u00eb caktuara. Ashtu si n\u00eb debug-er\u00ebt tradicional\u00eb, Debugger-i i Transformer-it mb\u00ebshtet navigimin hap pas hapi n\u00eb daljen e modeleve, ndjekjen dhe kapjen e aktivitetit t\u00eb caktuar. N\u00eb m\u00ebnyr\u00eb t\u00eb p\u00ebrgjithshme, Debugger-i i Transformer-it lejon t\u00eb kuptosh pse nj\u00eb model gjuh\u00ebsor jep nj\u00eb token t\u00eb caktuar si p\u00ebrgjigje ndaj nj\u00eb k\u00ebrkese t\u00eb caktuar n\u00eb vend t\u00eb nj\u00eb tokeni tjet\u00ebr, ose pse modeli i kushton m\u00eb shum\u00eb r\u00ebnd\u00ebsi token\u00ebve t\u00eb caktuar n\u00eb k\u00ebrkes\u00eb. Kodi \u00ebsht\u00eb shkruar n\u00eb gjuh\u00ebn Python dhe shp\u00ebrndahet n\u00ebn licenc\u00ebn MIT. <\/p>\n
Pjes\u00ebt p\u00ebrfshijn\u00eb komponent\u00ebt si vijon: <\/p>\n
\n
Neuron viewer — naviguesi n\u00ebp\u00ebr komponent\u00ebt e ve\u00e7ant\u00eb t\u00eb modelit, si\u00e7 jan\u00eb neuronet MLP, fokuset e v\u00ebmendjes dhe p\u00ebrfaq\u00ebsimet e fshehura t\u00eb auto-enkoduesve.\n
Activation server — backend server q\u00eb komunikon me modelin e analizuar dhe nxjerr t\u00eb dh\u00ebna p\u00ebr debuggerrin.\n
Models — bibliotek\u00eb p\u00ebr nd\u00ebrveprimin me modelet gjuh\u00ebsore GPT-2 dhe auto-enkoduesit e p\u00ebrdorur, duke siguruar z\u00ebvend\u00ebsime p\u00ebr kapjen e aktivizimeve.\n
Shembuj t\u00eb grupeve t\u00eb t\u00eb dh\u00ebnave aktivizuese p\u00ebr neuronet MLP, v\u00ebmendjen dhe p\u00ebrfaq\u00ebsimet e fshehura t\u00eb koduesve t\u00eb vetvetish\u00ebm. <\/ul>\n