Jagoran Harshe AI

RoBERTa Horon Recipe

RoBERTa ya nuna cewa BERT ba a horar da shi sosai: ta hanyar daidaita girke-girke maimakon gine-gine, ya kafa sabbin bayanan tarihi.

2 min karatuAn sabunta ta ƙarshe

Dubawa

It is a masterclass in how training choices matter as much as model design.

Zurfafa nutsewa

RoBERTa (Ƙarfafa Ingantaccen Hanyar BERT), wanda Facebook AI ya fitar a cikin 2019, ya kiyaye gine-ginen BERT da gaske bai canza ba amma ya sake fasalin yadda aka horar da shi. Tawagar ta horar da dogon lokaci akan ƙarin bayanai (160GB na rubutu da na BERT's 16GB), sun yi amfani da batches mafi girma, kuma sun cire makasudin tsinkayar jumla na gaba na BERT bayan gano shi ba shi da amfani. Sun canza daga abin rufe fuska - inda aka rufe kalmomi iri ɗaya a kowane zamani - zuwa abin rufe fuska mai ƙarfi wanda ke sake rufewa duk lokacin da aka ga jerin, kuma suna amfani da tokenizer na BPE-matakin byte. Tare da waɗannan canje-canjen kadai, RoBERTa ya zarce BERT kuma ya dace ko ya doke sabbin samfura kamar XLNet akan GLUE, SQuAD, da RACE, yana tabbatar da cewa horarwar da ta dace na iya yin hamayya da ƙirƙira ƙira.

Fahimtar Fasaha

Maɓallin maɓalli na RoBERTa sun kasance ma'auni da sarrafa bayanai, ba sabon yadudduka ba. Tsayawa abin rufe fuska yana haifar da sabon salo na abin rufe fuska akan tashi don kowane misalin horo, yana fallasa ƙirar zuwa ƙarin bambance-bambancen hasashen hasashen. Zubar da tsinkayar jimla ta gaba da horarwa akan jimloli masu tsayi masu tsayi (cikakken jimla') sun sauƙaƙa makasudin. Haɗe tare da manyan nau'ikan tsari (har zuwa jeri 8K), ingantaccen jadawalin koyo, da kuma babban littafin BookCorpus + CC-News + OpenWebText + Labarun corpus, waɗannan zaɓin sun ɗaga daidaito a ƙasa sosai.

Dabarun Tasiri

Gudu da sikelin

Gudun aikin harshe na iya tafiya da sauri ba tare da sadaukar da daidaito ba.

Shiga ku isa

Yana faɗaɗa damar shiga cikin harsuna da salon sadarwa.

Shawarwari masu haske

Ƙungiyoyi za su iya ciyar da ƙarin lokaci akan hukunci yayin da aiki da kai ke sarrafa maimaitawa.

Makomar girkin Horon RoBERTa

Darasi mai ɗorewa na RoBERTa - cewa a hankali bayanai, ma'auni, da daidaitawa na hyperparameter na iya yin nauyi fiye da tweaks na gine-gine - siffa yadda filin ke fuskantar horo. Ya kasance babban amfani da ko'ina, abin dogaro da kashin bayan bayanan sirri don rarrabuwa, dawo da aiki, da kyawawan ayyuka, kuma bambance-bambancen harsuna da yawa kamar XLM-R sun tsawaita girke-girke a cikin harsuna 100. Yayin da tunanin shari'a ya balaga, falsafar RoBERTa na 'horo mafi kyau, ba kawai manyan gine-gine' na ci gaba da sanar da ingantaccen haɓaka samfuri.

Aiwatar da Gaskiyar Duniya

Kyakkyawan daidaitawa RoBERTa don nazarin ji, gano guba, da daidaita abun ciki

Yin aiki azaman mai ƙarfi mai rikodin ma'anar bincike da ƙirar jumla

Ƙarfafa NLP na harsuna da yawa ta hanyar bambance-bambancen XLM-RoBERTa a cikin harsuna 100

Yin aiki azaman babban madaidaicin tushe akan ma'auni na glue, SQuAD, da RACE

Hatsari & Tsare-tsare

Abubuwan da aka ruɗe suna iya shigar da rahotanni cikin nutsuwa, kwararar tallafi, ko abubuwan bincike.

Hankali na gaggawa na iya ƙirƙirar sakamako mara daidaituwa a cikin buƙatun iri ɗaya.

Za a iya fallasa bayanan rubutu mai ma'ana idan ikon samun dama yana da rauni.

Taswirar Hanya

1

Ƙayyade tsarin fitarwa, sautin, da ma'auni masu inganci kafin fitowa.

2

Amsa a ƙasa tare da amintattun tushe a duk lokacin da daidaito ya shafi mahimmanci.

3

Ajiye wurin binciken ɗan adam don abubuwan da ake samu masu girma.

4

Bibiyar tsarin gazawar kuma sake horar da tsokaci ko tafiyar aiki akai-akai.

Ci gaba da Bincike

Free newsletter

Get the daily AI briefing

Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.

One email each weekday. Unsubscribe in one click. We never sell or share your address.

Test yourself

Take the RoBERTa Training Recipe quiz

Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.

Fara tambayoyi

Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation

Jagora na gaba

Horon Hasashen Multi-Token

Tambayoyin da ake yawan yi

What is RoBERTa Training Recipe?

RoBERTa ya nuna cewa BERT ba a horar da shi sosai: ta hanyar daidaita girke-girke maimakon gine-gine, ya kafa sabbin bayanan tarihi. Babban aji ne kan yadda zaɓin horarwa ke da mahimmanci kamar ƙirar ƙira.

Menene ainihin fahimtar RoBERTa game da ainihin BERT?

RoBERTa ya nuna cewa BERT ba a horar da shi, kuma ƙarin bayanai da dogon horo sun inganta sakamako sosai.

Wace manufar BERT RoBERTa ta cire?

RoBERTa ya sami tsinkayar jimla ta gaba ba ta da amfani kuma ta watsar da shi, horarwa kawai tare da abin rufe fuska.

Menene tsayayyen abin rufe fuska a cikin RoBERTa?

Ba kamar abin rufe fuska na BERT ba, RoBERTa yana sake rufe jerin abubuwan rufe fuska a hankali, yana fallasa samfurin ga maƙasudin hasashen iri iri.

Yaya aka kwatanta bayanan horon RoBERTa da na BERT?

RoBERTa ya horar akan kusan 160GB na rubutu (BookCorpus, CC-News, OpenWebText, Labarun) tare da BERT's kusan 16GB.

Wace kungiya ce ta saki RoBERTa?

Facebook AI ne ya gabatar da RoBERTa (yanzu Meta AI) a cikin 2019.