Modèlu kodu kontekst bu gudd IA luxus
Magic IA dafa tabax ay modelu defar kode frontiere yuñ wuutale ci palanteer yu gudd lool, may model bi mu jàng codebase bi yépp benn yoon.
Résumé
It matters because software understanding depends on context, and a model that can hold millions of lines in memory can reason about a whole project rather than one file.
Plongeur bu xóot
Magic AI ab ndoorte la buy defar ab ingenieur losisel IA te baña yam ci ab jumtukaay buy yeggali boppam. Li gëna am solo ci liggéeyam mooy LTM (Long-Term Memory) famiy model, boole ci LTM-2-mini, bi liggéeyukaay bi wax ni dafay jàppale palanteer yu yegg ba 100 milioŋ ciy token - lu tollu ci 10 milioŋ ci ligne kode wala ay junni téere yuñ tëye ci muy tekki benn yoon. Ci atum 2024 Magic yëglewoon na jàppante bu mag ak Cloud ngir tabax ay super ordinatër ci kaw jumtukaayi Nvidia ba noppi dajale téemeeri milioŋ ciy dolaar, ak ay jàppalekat yu ci melni Eric Schmidt. Ngir natt yokkute bi weesu benchmarks yu yomb-memoriser, Magic sos HashHop, ab evaluation buy jëfandikoo chaînes hash yu bari yu benn model mënul fàttaliku ci tàggat, forse récupération bu yàgg.
Gis-gis xarala
Transformatër yiñ miin dañuy bàyyi seen xel ci eskaalu quadratically ak guddaayu toppalante bi, moo tax 100-million-token contexte seer lool ak pexe yu yomb. Magic dafa wax ni algorithm bimu am LTM-2-mini dimension sequence moo gëna xéewale token bu nekk ci anam wu mel nii, loolu mooy tax ñu mëna am contexte bu gudd te yomb. HashHop benchmark dafay wecci ay tegtali semantik ak ñaari hash yu bari, yu mënul kompresse, kon benn anam bi ñu mëna tontu mooy jëlaat ak chaine xibaar ci palanteer bi yépp - test bu gëna tar ci mën mënu contexte bu yàgg.
njeextalu pexe
Pexem jaaykat
Kartu yoonu jaaykat yi ñooy wane man-man yi sa ekip mëna tabax ci kanam.
Njëgg ak budget
Anamu jënd ak jaay ak tànneefi dugal dañu am njeexital ci njëg ak risk ci diir bu xawa yàgg.
Risk ak kaaraange
Li liggéeyukaay bi di ñaax mooy tëral ni produit bi di doxee, kaaraange gi ak ubbeeku gi.
Ëlëgu IA lu am solo
Sudee model yi mën nañu tëye ak xalaat ci kaw codebase bi yépp, assistant IA yi dina ñu joge ci digle ay snippet ngir def ay refactor ci projet bi yépp, topp ay bug ci fichier yu bari, ak samp ay man-man yuy laal ay module yu bari. Jafe-jafe yi ubbeeku mooy tëye inference ultra-long-context gaaw te yomb te firndeel ni model bi dafay jëfandikoo contexte bu sori dëgg moo gën ñu ko baña bàyyi xel. Xaarandi contexte bu gudd ak def liggéeyu agence ngir dajaloo ci sistem yuy def ni ay naataango yu dëggu ci ingenieur losisel.
Doxal ci àdduna dëgg
Yab ab dencukaay bu yaatu lëmm suko defee model bi mëna tontu laaj yi jëm ci ni module yu sori yi di jaxasoo.
Def ab refactor bu yaatu ci projet bi, fu coppite ci interfaasu benn fichier di tasaare ci anam wu jaar yoon ci codebase bi yépp.
Topp benn bug bu sabab bi laal fichier yu bari ci xalaat ci contexte bi yépp benn yoon ci barabu fichier-ci-fichier.
Onboarding ci codebase buñu xamul ci laaj model bi mu tënk architecture bi jëfandikoo source bi yépp ni contexte.
Risk yi ak balustrade yi
Koom-koomu ubbite mën na raw stabilite ci def liggéeyu defar dëgg.
Njëg yi ci API wala coppite ci sàrt yi mën nañu dindi xalaat yi ci guddi gi.
Dependence ci benn jaaykat dafay yokk njëgu tëjug ak migraasioŋ.
Roadmap ngir samp gi
Saytu sa fournisseur yi nga jëfandikoo sa liggéey ak say done.
Xoolaat mbir yu nëbbu, kaaraange ak sàrti yoon balaa ngay boole.
Fexe am palaŋu fallback ci model yi wala jaaykat yi.
Xool notu génne yi suko defee coppite yi ci kàrtu yoon du jaaxal ekip yi.
Weyal di banneexu
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 Magic AI Long-Context Code Models quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Gis bi ci topp
Protokolu muy misaal
Laaj yi ñuy faral di laaj
What is Magic AI Long-Context Code Models?
Magic IA dafa tabax ay modelu defar kode frontiere yuñ wuutale ci palanteer yu gudd lool, may model bi mu jàng codebase bi yépp benn yoon. Dafa am solo ndax xam-xam losisel mingi aju ci muy tekki, te model bu mëna tëye ay milioŋ ciy ligne ci memory mën na xalaat ci projet bi yépp moo gën benn fichier.
Lan mooy Magic AI di njëkka jéema tabax?
Magic dafa bëgga defar ab insiñër losisel IA, te baña yam ci ab jumtukaay buy yeggali boppam.
Ban dayo palanteer la Magic wax ci model LTM yi?
Magic yëgle na ni dafay jàppale palanteer yu yegg ba 100 milioŋ ciy jeton, lu tollu ci 10 milioŋ ci liiñ kode.
Luy hashhop?
HashHop dafay jëfandikoo ay hash yu bari te kenn mënu leen kompresse, moo tax model bi dafa wara am ci dëgg-dëgg ba noppi boole ay leeral ci contexte bi yépp.
Lan moo waral bàyyi xel ci transformatër bu xamadi nekk jafe-jafe ci 100 milioŋ ciy token?
Njëg li ñuy fay ci bàyyi xel dafay gëna yokk luñuy gëna yokk guddaayu toppalante bi, kon contexte yu gudd lool dañu seer lool te amul algorithm yu gëna baax.
Ban liggéeyukaay la Magic jëflante ak moom ci 2024 ngir defar ay super ordinatër?
Magic announced a partnership with Google Cloud to build supercomputers on Nvidia hardware.