Overview
Instruction tuning inhanho yekudzidzisa inoshandura nyoro-predictor kuita modhi inotevedzera mirairo yakaita se'pfupisa izvi' kana 'nyora mhinduro ine hunhu.' Izvo ndizvo zvinoita kuti base modhi inzwe inobatsira uye inotungamira.
Mirayiridzo Tuning chikamu chemutauro-AI stack inoshandiswa kuverenga, kugadzira, kuronga, uye kushandura zvinyorwa uye kutaura pamwero.
Deep Dive
Modhi yemutauro wepasi inodzidziswa kungofembera chiratidzo chinotevera pamavara ewebhu, saka kana ukanyora mubvunzo unogona kungoenderera uine mimwe mibvunzo pane kupindura. Instruction tuning inogadzirisa izvi. Iyo imhando yekutariswa kwakanyatso kurongeka: modhi inodzidziswa pazviviri zviviri zve (mirayiridzo, mhinduro yakanaka) inovhara zviuru zvemabasa - kududzira, kupfupisa, kurongedza, Q&A, kukodha, nezvimwe. Nekuona iyo imwechete yekuraira-uye-inobatsira-mhinduro pateni kakawanda, modhi inodzidza maitiro akajairwa e'ita izvo zvinokumbirwa nemushandisi,' uye izvi zvinokonzeresa kune mirairo yaasina kumboona mukudzidziswa. Nzira iyi yakavambwa muna 2021 nebasa rakaita seFLAN, T0, uye Natural Instructions, uye yaive pakati pe OpenAI's InstructGPT, iyo yakanyatso gadzirisa GPT-3 pane yakasarudzika seti yemirairo yekurudziro. Ndiwo hwaro huzhinji hwevabatsiri vekutaura vanovakirwa pahuri.
Technical Insight
Mechanical, kuraira kurongedza kudzidza kwakatariswa kwakajairwa: kuderedza mutsauko uripo pakati peiyo modhi yakafanotaurwa tokens nemhinduro yereferenzi, nemagradient achivandudza huremu. Yakasiyana neRLHF (kusimbisa kudzidza kubva mumhinduro dzevanhu), iyo inouya mushure uye inogadzirisa zvido zvevanhu vachishandisa mubairo modhi. Resipi yenguva dzose yakaturikidzana: pretrain, ipapo kuraira-tune (SFT) kudzidzisa basa-kutevera, wozosarudza RLHF kukwenenzvera toni, kubatsira, uye kuchengeteka. Kusiyana-siyana kwedata kunonyanya kukosha kupfuura sheer vhoriyamu - yakafara basa kufukidzwa inofambisa iyo generalization.
Mastering Instruction Tuning
Instruction tuning inhanho yekudzidzisa inoshandura nyoro-predictor kuita modhi inotevedzera mirairo yakaita se'pfupisa izvi' kana 'nyora mhinduro ine hunhu.' Izvo ndizvo zvinoita kuti base modhi inzwe inobatsira uye inotungamira. Mirayiridzo Tuning chikamu chemutauro-AI stack inoshandiswa kuverenga, kugadzira, kuronga, uye kushandura zvinyorwa uye kutaura pamwero. Kuti uvake kunzwisisa kwakadzama, tora Instruction Tuning semuenzaniso wekushandisa, kwete chinhu chimwe chete: tsanangura zvaunoda, kujekesa fungidziro, uye patsanura izvo zvinogona kuitwa nehurongwa hwakavimbika kubva kune zvichiri kuda kutonga kwenyanzvi.
Mukuita, zvikwata zvakasimba zvinoshandisa Instruction Tuning dhizaini inokurudzira, kudzoreredza, uye kuongorora zvishwe seimwe yakabatanidzwa yekutaurirana system. Ivo vanonyora zvakajeka maitiro ebudiriro, bvunzo vachipokana ne data rechokwadi uye mafambiro ebasa, uye iterate zvichibva pane zvakacherechedzwa maitiro ekutadza kwete kuhwina-nguva imwe chete yebhenji. Apa ndipo apo kunzwisisa kwe theoretical kunoshanduka kuve kugona kwakasimba pane chigadzirwa, mutemo, uye mashandiro.
Mutauro workflows inogona kufamba nekukurumidza pasina kupira kuenderana. Panguva imwecheteyo, chokwadi cheHallucified chinogona kupinda chinyararire mishumo, kuyerera kwetsigiro, kana kutsvagisa zvinobuda. Nzira yakatsiga ndeyekubatanidza kukurumidza kuyedza nekutonga: mhanyisa vatyairi vendege, tora humbowo, buritsa matanda esarudzo, uye urambe uchivandudza chengetedzo semaitiro emuenzaniso, zvinotarisirwa nemushandisi, uye zvinodikanwa zvekutonga.
Strategic Impact
Mutauro workflows inogona kufamba nekukurumidza pasina kupira kuenderana.
Mutauro workflows inogona kufamba nekukurumidza pasina kupira kuenderana. Mukutumirwa kwemhando yepamusoro, izvi zvinoshandurirwa kuita mitemo inoyerwa yekushanda, miganhu yevaridzi, uye tsika dzekudzokorora dzinodzokororwa kuitira kuti zvikwata zvikwire kuvimba pane kukwidza kusajeka.
Inopamhidzira kupinda mumitauro yese nemataera ekutaurirana.
Inopamhidzira kupinda mumitauro yese nemataera ekutaurirana. Mukutumirwa kwemhando yepamusoro, izvi zvinoshandurirwa kuita mitemo inoyerwa yekushanda, miganhu yevaridzi, uye tsika dzekudzokorora dzinodzokororwa kuitira kuti zvikwata zvikwire kuvimba pane kukwidza kusajeka.
Zvikwata zvinogona kupedza nguva yakawanda pakutonga uku otomatiki ichibata kudzokorora.
Zvikwata zvinogona kupedza nguva yakawanda pakutonga uku otomatiki ichibata kudzokorora. Mukutumirwa kwemhando yepamusoro, izvi zvinoshandurirwa kuita mitemo inoyerwa yekushanda, miganhu yevaridzi, uye tsika dzekudzokorora dzinodzokororwa kuitira kuti zvikwata zvikwire kuvimba pane kukwidza kusajeka.
Real-World Implementation
Kushandura iyo base GPT-style modhi kuita mubatsiri wekutaura anopindura mibvunzo pachinzvimbo chekuita maungira
FLAN-T5, yakanyatsorongedzerwa pamabasa mazhinji kuitira kuti ikwanise kutevedzera mirairo yayasina kumbodzidziswa nayo
InstructGPT, uko GPT-3 yaive yakarongedzerwa pane curated kukurudzira kuburitsa zvakanyanya kubatsira mhinduro.
Kuvaka mubatsiri wekambani yemukati nekugadzirisa zvakanaka pamirayiridzo-yemhinduro mbiri dzakanyorwa nerutsigiro uye zvikwata zvemutemo
Maitiro Ekuita
Mirayiridzo Tuning mukuita
Kushandura iyo base GPT-style modhi kuita mubatsiri wekutaura anopindura mibvunzo pachinzvimbo chekuita maungira.
Kushandura iyo base GPT-style modhi kuita mubatsiri wekutaura anopindura mibvunzo pachinzvimbo chekuipindura Matimu anowanzo kuwana mhedzisiro iri nani kana achinge atsanangura emhando yepamusoro kumberi, chengetedza nzira yekukwira kwevanhu yemakesi ekumucheto, uye kuteedzera zvese zvakawanikwa zvechigadzirwa nemitengo yekukanganisa nekufamba kwenguva.
Mirayiridzo Tuning mukuita
FLAN-T5, yakanyatsorongedzerwa pamabasa mazhinji kuitira kuti ikwanise kutevedzera mirairo iyo isina kumbobvira yanyatsodzidziswa pairi.
FLAN-T5, yakanyatsorongedzerwa mumabasa mazhinji kuti iteedzere mirairo iyo isina kumbobvira yadzidziswa zvakajeka paTimu kazhinji inowana mhedzisiro iri nani painotsanangura hunhu hwepamberi, chengetedza nzira yekukwira kwevanhu yemakesi ekumucheto, uye kuteedzera zvese zvakawanikwa zvechigadzirwa nemitengo yekukanganisa nekufamba kwenguva.
Mirayiridzo Tuning mukuita
InstructGPT, uko GPT-3 yaive yakarongedzerwa-yakarongedzwa pane curated kukurudzira kuburitsa zvakanyanya kubatsira mhinduro.
InstructGPT, uko GPT-3 yaive yakarongedzerwa pane zvakavharwa kuti ibudise mhinduro dzinobatsira Matimu anowanzo kuwana mibairo iri nani kana achinge atsanangura hunhu hwepamberi, chengetedza nzira yekukwira kwevanhu yemakesi emupendero, uye kuteedzera zvese zvakawanikwa zvechigadzirwa nemitengo yekukanganisa nekufamba kwenguva.
Mirayiridzo Tuning mukuita
Kuvaka mubatsiri wekambani yemukati nekugadzirisa zvakanaka pamirayiridzo-yemhinduro mbiri dzakanyorwa nerutsigiro uye zvikwata zvemutemo.
Kuvaka mubatsiri wekambani yemukati nekugadzirisa zvakanaka pamirairo-mhinduro mbiri dzakanyorwa nerutsigiro uye zvikwata zvepamutemo Matimu anowanzo kuwana mhedzisiro iri nani kana achinge atsanangura emhando yepamusoro kumberi, chengetedza nzira yekukwira kwevanhu yemakesi ekumucheto, uye kuteedzera zvese zvakawanikwa zvechigadzirwa uye mutengo wekukanganisa nekufamba kwenguva.
Njodzi & Guardrails
Chokwadi chehuroyi chinogona kupinda chinyararire mishumo, kuyerera kwetsigiro, kana tsvakiridzo.
Kunzwa nekukasira kunogona kugadzira mhedzisiro isingaenderane pane zvikumbiro zvakafanana.
Sensitive text data inogona kuburitswa kana zvidhiraivho zvisina kusimba.
Implementation Roadmap
Tsanangura chimiro chekubuda, toni, uye mhando zviyero usati waburitsa.
Tsanangura chimiro chekubuda, toni, uye mhando zviyero usati waburitsa. Bata nhanho yega yega segedhi rehumbowo: kana maitiro asina kusangana, imbomira kuburitsa, vhara gaka, uye wobva wawedzera kushandiswa.
Mhinduro dzepasi neakavimbika masosi pese pazvine basa.
Mhinduro dzepasi neakavimbika masosi pese pazvine basa. Bata nhanho yega yega segedhi rehumbowo: kana maitiro asina kusangana, imbomira kuburitsa, vhara gaka, uye wobva wawedzera kushandiswa.
Chengetedza ongororo yekuongorora yemunhu kune yakakwira-stake zvinobuda.
Chengetedza ongororo yekuongorora yemunhu kune yakakwira-stake zvinobuda. Bata nhanho yega yega segedhi rehumbowo: kana maitiro asina kusangana, imbomira kuburitsa, vhara gaka, uye wobva wawedzera kushandiswa.
Tevera maitiro ekutadza uye dzidzisazve kukurudzira kana mafambiro ebasa nguva nenguva.
Tevera maitiro ekutadza uye dzidzisazve kukurudzira kana mafambiro ebasa nguva nenguva. Bata nhanho yega yega segedhi rehumbowo: kana maitiro asina kusangana, imbomira kuburitsa, vhara gaka, uye wobva wawedzera kushandiswa.