Résumé
Liquid IA ab tabax la bu juge ci MIT buy tabax ay modelu fondaasioŋ buy ndox (LFMs) yuy bàyyi Transformer yiñ miin ngir tabax ay sistem yu am doole. Luñu bëgga mooy model yu ndaw, gaaw, sakkanal mémoire, yuy dox ci telefon yi ak aparey yi te duñu yàq kalite bu bari.
Liquid IA ak Liquid Fondation Models ñu ngi koy gëna xamee ci wàllu pexe, jëfandikoo model, dogal ci platform, ak jàppante ci ecosystem.
Plongeur bu xóot
Ramin Hasani, Mathias Lechner, Alexander Amini ak Daniela Rus ñoo sos juntuwaay yu bees yi ci 2023, ekipu MIT CSAIL bi ci ginnaaw 'reso neuronal yu ndoxe yi.' Ñoom ñu ngi bawoo ci jàngat ci saxaaru nematode C. elegans, moom mi yuur gi am 302 neuron yu ndaw yu inspiré ay reso Liquid Time-constant (LTC) fu neuron bu nekk di soppeeku ci diir bu gàtt jaaraleko ci equation différentiel. Royuwaayi njaay yu Liquid, Royuwaayi Fondaasioŋ Liquid (LFM-1B, 3B, 40B), dañuy yaatal xalaat boobu ginaaw Transformatër yi. Benn ci màndarga yi gëna fës mooy emprent mémoire bu sax bi contexte bi di màgg, wuute na ak Transformers yi seen xel di cache balloon ak guddaayi sekans. Ci 2024 liggéeyukaay bi dajale Series A bu mag (ñu xamle ni lu tollu ci 250 milioŋ ciy dolaar) ci ganaw ga mu génne LFM2, ñu defar ko ngir ñu mëna ko jëfandikoo ci ordinatër portable, telefon ak oto.
Gis-gis xarala
Transformatër yi dañuy denc cache valeur bu am caabi buy màgg lineairement ak guddaayu dugal bi, kon contexte yu gudd yi dañuy lekk mémoire bi. Lu moy loolu LFM yi dañuy jëfandikoo 'liquid' yunit ordinatër yuñ tabax ci espace-state buñ yamale ak operatëri system dynamique yuy kompresse xibaar yi weesu ci stade recurrent bu am dayo fixe. Equation à temps continue yi ñooy leeral xayma yi, te seeni paramètre (lu melni constants à temps) dañu méngoo ak li ñuy dugal, loolu mooy tax model bi mëna jëfandikoo ay sequence yu gudd ak memory bu plat ak latency buñ mëna seentu, te loolu mooy gëna baax ci hardware yu am resource yu néew.
Xam IA bu ndoxe ak xeetu fondasioŋ bu ndoxe
Liquid IA ab tabax la bu juge ci MIT buy tabax ay modelu fondaasioŋ buy ndox (LFMs) yuy bàyyi Transformer yiñ miin ngir tabax ay sistem yu am doole. Luñu bëgga mooy model yu ndaw, gaaw, sakkanal mémoire, yuy dox ci telefon yi ak aparey yi te duñu yàq kalite bu bari. Liquid IA ak Liquid Fondation Models ñu ngi koy gëna xamee ci wàllu pexe, jëfandikoo model, dogal ci platform, ak jàppante ci ecosystem. Ngir tabax xam-xam bu xóot, jàppal Liquid AI ak Liquid Foundation Models ni xeetu liggéey, du benn man-man: leeral njariñ yi nga bëgg, leeral xalaat yi, ak tàqale li sistem bi mëna def ci anam wu wóor ak li ba leegi soxla àtteb kàngam.
Ci jëf, ekip yu am doole yiy jëfandikoo Liquid AI ak Liquid Foundation Models dañuy jàngat pexem jaaykat bi, wóorum kàrtu yoon wi, ak risku tëjug balaa ñuy jël dogal. Dañuy bind kritër yu leer ngir am ndam, natt leen ci done yu dëggu ak def liggéey, ba noppi ñu baamtu ci anamu ñàkka mëna seetlu, du ci benn yoon benchmark wins. Mooy barab bi xam-xam theorie bi di soppiku nekk kàttan buy yàgg ci produit yi, ci politik yi ak ci liggéey yi.
Kartu yoonu jaaykat yi ñooy wane man-man yi sa ekip mëna tabax ci kanam. Ci jamano jooju, annonce lansma yi mën nañu raw stabilite ci liggéeyu production dëgg. Xeetu jëf bi gëna dëgër mooy boole gaawaayu jàngat ak disipline nguur: doxal pilote, jàpp firnde, siiwal dogal yi, ak wéy di yeesal kaaraange gi ci anam wi ñuy doxalee, li jëfandikukat bi di xaar, ak sàrti sàrt yi di jëm kanam.
njeextalu pexe
Kartu yoonu jaaykat yi ñooy wane man-man yi sa ekip mëna tabax ci kanam.
Kartu yoonu jaaykat yi ñooy wane man-man yi sa ekip mëna tabax ci kanam. Ci jëfandikoo yu am kalite bu kawe, loolu dañu koy tekki ci sàrti liggéey yuñ mëna natt, ay peggu boroom, ak ay xew-xewu xoolaat yu bari suko defee ekip yi mëna yokk wóolu seen bopp ci barabu yokk lu jaxasoo.
Anamu jënd ak jaay ak tànneefi dugal dañu am njeexital ci njëg ak risk ci diir bu xawa yàgg.
Anamu jënd ak jaay ak tànneefi dugal dañu am njeexital ci njëg ak risk ci diir bu xawa yàgg. Ci jëfandikoo yu am kalite bu kawe, loolu dañu koy tekki ci sàrti liggéey yuñ mëna natt, ay peggu boroom, ak ay xew-xewu xoolaat yu bari suko defee ekip yi mëna yokk wóolu seen bopp ci barabu yokk lu jaxasoo.
Li liggéeyukaay bi di ñaax mooy tëral ni produit bi di doxee, kaaraange gi ak ubbeeku gi.
Li liggéeyukaay bi di ñaax mooy tëral ni produit bi di doxee, kaaraange gi ak ubbeeku gi. Ci jëfandikoo yu am kalite bu kawe, loolu dañu koy tekki ci sàrti liggéey yuñ mëna natt, ay peggu boroom, ak ay xew-xewu xoolaat yu bari suko defee ekip yi mëna yokk wóolu seen bopp ci barabu yokk lu jaxasoo.
Doxal ci àdduna dëgg
Doxal ab assistant chat bu mëna jëfandikoo ci telefon bu xarañ ngir jëfandikoo ci anam wu yomb
Dugalal xam-xam làkk bu woyof ci biir oto ngir doxal kàddu te du am tukkib cloud
Liggéeyukaay dokimaa wala log yu gudd lool ci ordinatër portable fu cache memory Transformer bi di yaatu lool
Robotik yu am doole ak aparey IoT yu C. elegans-inspired reso liquid yi gëna am doole ci doxal bu wéy
Modèlu jëfandikoo
IA bu ndoxe ak xeetu fondasioŋ bu ndoxe ci jëf
Doxal ab assistant chat bu mëna def te nekk ci offline ci telefon bu xarañ ngir jëfandikoo ci anam wu yomb.
Doxal ab assistant chat bu mëna jëfandikoo ci benn telefon bu xarañ ngir jëfandikoo bu am njariñ. Ekip yi dañuy faral di am njariñ yu gëna baax suñu joxe ay tegtal yu baax ci kanam, tëye yoonu escalation nit ngir jafe-jafe yi, ba noppi topp njariñu produit ak njëgu njuumte ci diir bu gàtt.
IA bu ndoxe ak xeetu fondasioŋ bu ndoxe ci jëf
Dugalal xam-xam làkk bu woyof ci biir oto ngir mëna doxal baat te du am cloud buy dem ak dikk.
Dafay dugal xam-xam làkk bu woyof ci oto yi ngir doxal baat te amul cloud round-trips Ekip yi dañuy faral di am njariñ yu gëna baax suñu joxee threshold yu baax ci kanam, tëye yoonu escalation nit ngir jafe-jafe yi, ba noppi topp njariñu produit yi ak njëgu njuumte yi ci diir bi.
IA bu ndoxe ak xeetu fondasioŋ bu ndoxe ci jëf
Liggéeyukaay dokimaa yu gudd lool wala log ci ordinatër portable fu cache memory Transformer bi di yaatu lool.
Liggéeyukaay dokimaa yu gudd lool wala log ci ordinatër portable fu cache memory Transformer di gëna yaatu. Ekip yi dañuy faral di am njariñ yu gëna baax suñu joxee thresholds yu baax ci kanam, tëye yoonu escalation nit ngir jafe-jafe yi, ba noppi topp njariñu produit ak njëgu njuumte ci diir bi.
IA bu ndoxe ak xeetu fondasioŋ bu ndoxe ci jëf
Robotik yu am doole ak aparey IoT fu reso liquid yu C. elegans yi gëna am doole ci doxal bu wéy.
Robotik yu am doole ak aparey IoT yu am doole C. elegans-inspired liquid networks excel ci njiitu wéyal Ekip yi dañuy faral di am njariñ yu gëna baax suñu leeralee threshold yu kalite ci kanam, tëye yoonu escalation nit ngir jafe-jafe yi, ak topp njuréefi produit ak njëgu njuumte ci diir bi.
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.
Saytu sa fournisseur yi nga jëfandikoo sa liggéey ak say done. Japp jéego bu nekk ni buntu firnde: sudee mattul kritër yi, noppali génne gi, tëj bërëb bi, ba noppi nga yaatal jëfandikoo gi.
Xoolaat mbir yu nëbbu, kaaraange ak sàrti yoon balaa ngay boole.
Xoolaat mbir yu nëbbu, kaaraange ak sàrti yoon balaa ngay boole. Japp jéego bu nekk ni buntu firnde: sudee mattul kritër yi, noppali génne gi, tëj bërëb bi, ba noppi nga yaatal jëfandikoo gi.
Fexe am palaŋu fallback ci model yi wala jaaykat yi.
Fexe am palaŋu fallback ci model yi wala jaaykat yi. Japp jéego bu nekk ni buntu firnde: sudee mattul kritër yi, noppali génne gi, tëj bërëb bi, ba noppi nga yaatal jëfandikoo gi.
Xool notu génne yi suko defee coppite yi ci kàrtu yoon du jaaxal ekip yi.
Xool notu génne yi suko defee coppite yi ci kàrtu yoon du jaaxal ekip yi. Japp jéego bu nekk ni buntu firnde: sudee mattul kritër yi, noppali génne gi, tëj bërëb bi, ba noppi nga yaatal jëfandikoo gi.