Sakana IA dafay boole xeetu jëm kanam
Sakana IA labo la bu nekk ci Tokyo, muy jëfandikoo ay pexe yu juge ci nature bi ci IA, rawatina jëfandikoo ay algorithm yuy jëm kanam ngir boole ay xeetu ubbeeku yu fi nekk ci yeneen yu bees te gëna baax.
Résumé
Instead of training from scratch, it 'breeds' models by automatically combining their strengths.
Plongeur bu xóot
Sakana AI was founded in 2023 by Llion Jones, a co-author of the original 'Attention Is All You Need' Transformer paper, and David Ha, formerly of Google Brain. The name means 'fish' in Japanese, reflecting a philosophy inspired by schools and swarms: many small, collective agents rather than one giant model. Its breakthrough technique, Evolutionary Model Merging, uses evolutionary search to discover how to combine the weights and layers of multiple pretrained open-source models. The algorithm explores thousands of merge recipes, keeping combinations that score well on target tasks. Sakana used this to create capable Japanese-language and Japanese math and vision models by merging existing models, at a tiny fraction of the cost of training new ones. The company also produced the 'AI Scientist,' a system that attempts to automate research itself.
Gis-gis xarala
Modèlu boole dafay boole parametru reso yuñ tàggat bu baax. Sakana dafay jëm kanam boole ci ñaari barab benn yoon: barabu parametre (ni ñuy pondéree ak interpoler poids model bu nekk, couche par couche) ak espace flow de données (ban couche lañu wara jëlee ci model yi ak ci ban yoon). Algorithm buy jëm kanam dafay tànn rëset yuñ mëna tànn, jàngat leen ci benn référence, ba noppi tànnee ba noppi soppali li gëna baax, baamtu ci ibrid yu am doole te amul benn tàggat bu lalu ci gradient.
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 boole xeetu jëm kanam bu Sakana IA
Mbootaayu jëm kanam gi dafay wane ëlëg fu ñuy boole ay xeetu model yu bees ci bibliotek bu mag bu model yu ubbeeku ci njëg yu yomb, di demokratise jëfandikoo gi ginaaw laboratuwaar yi am budget ordinatër yu bari. Buñu ko boole ak 'Gëstukat IA' bu Sakana, gis-gis bi gëna gudd mooy sistem IA yiy jàppale nit ñi ñu gis seen yokkute. Laajte yu ubbeeku yi ñooy moytu model yuñ boole yuy jël ay bug wala ay njuumte, ak ndax seetlu jëm kanam gi dafay yegg ba ci performance frontiere bi moo gën ñuy gëna spesialise model yi fi nekk.
Doxal ci àdduna dëgg
Defar ab xeetu lakk bu dëgër bu mëna làkk japonais ci boole xeetu làkku angale ak japonais bu ubbeeku te doo tàggataat
Tabax ab xeetu xalaat ci math japonais ci jëm kanam ci boole ay xeetu math yuñ jagleel
Defar xeetu làkku gis-gis buy jëfandikoo bind japonais ci biir nataal jaaraleko ci boole domen yu bari
Bayyi mbootaay yu ndaw yi ñu dajale ay xeetu liggéey yu yomb ci poid yu ubbeeku yi ci plaasu tàggat ci noonu
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 Sakana AI Evolutionary Model Merging 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
boole model
Laaj yi ñuy faral di laaj
What is Sakana AI Evolutionary Model Merging?
Sakana IA labo la bu nekk ci Tokyo, muy jëfandikoo ay pexe yu juge ci nature bi ci IA, rawatina jëfandikoo ay algorithm yuy jëm kanam ngir boole ay xeetu ubbeeku yu fi nekk ci yeneen yu bees te gëna baax. Du tàggat ci noonu rek, waaye dafay 'yar' model yi ci boole seeni doole ci saasi.
Lan mooy boole xeetu jëm kanam bu Sakana IA?
Dafay jëm kanam ci rëset ngir boole xeetu diisaay ak diisaay yu ubbeeku yu bari yuñ tàggat bu njëkk, moo gën ñuy tàggataat.
Kan mooy benn ci ñi sos Sakana AI?
Llion Jones, kenn ci ñi bind téere bii tuddu 'Fàttaliku mooy li nga soxla', ñoom ak David Ha ñoo sos Sakana.
Lan la tur Sakana di tekki ak lumuy wane?
Sakana mooy jën; liggéeyukaay bi dafa taamu ay ndawu liggéey yu ndaw yu bari ci benn xeetu monolitik.
Sakana jëmmal xeetu boole ci ban ñaari espace?
Dafay seetlu ni ñuy pondérer/interpoler ay parametre ak ban couche lañu wara stack ak ci ban rang.
Lu weesu boole ay model, ban sistem bu am njariñ la Sakana dugal?
Sakana genne na 'AI Scientist', muy sistem buy jéema otomatise pàcci liggéeyu gëstu bi ci boppam.