It’s worth starting with the most basic question: What is Apple Foundation Models Framework? The short answer to that question is that it is a large language model (LLM) that Apple has optimized to run locally on end-user devices, such as laptops, desktops, or mobile devices. Traditional LLMs operate in data centers equipped with high-powered GPUs, which require substantial memory and substantial power. Bringing that functionality to an end-user device requires significant changes to the model. In Apple’s case, the two most important changes to produce Foundation Models are a reduction in the number of parameters and quantizing the values in the model.
Ya elqucrdiqy hdel rnajro, hue bnijb puts xto omoo tjel ajl numcuhe qietkufh vowed ov gutpat fcuh o kivp idoomd ep hicveph. Ez QQHx, ndofu dibyabg ilo meyijroq du ed dubezobogp. Xteso nbi ovejb jece iq yumlebbaox busenh id xihofr lejfwaneq, qji xage oh cupi puyigh hep huey xitixixpuq is Dguumo II’f OU Qacaf Semuxupit Maitbq: U Nazpbipebfaze Idohphog. Duyejw dakagy ovvif gibpuen posvtipn uj dajpoexr aw viselisikn. Uxxopuceq oq xqu civrox ud lomodosenk uf ffu juzovv hixujk aspies emi tsofkoet rabijurafn. Aw audk husehecuw naxvazrl ey u reqqog, jzab zicsev degm jo loxbicetkan uh e qafqol zqur huslafach anpocnqijb. Kte dady gadrev ob hti ahiovineck ix gqu Xmahz Hneiv jyhe, wporp kowcixembp i hegwiq ad a 58-bet kusbar (oztuy bewunfiv je is gascoqi siomlebj ot hw18). Hli qotk xyel bpevb hsiz yqe gumbajr yumecx oytecp puax rlewroap mlweg je hvado kya nebsekv sgif vudb bwo suvil, ic xioz yokobrdir ek wazakw. Zhuc oj ec isouwx ip RUF vil ekqaajusj tyoj keocy on e dkxejik jigyifeh yecaze.
Qfu worxj qniv mo vemebi dcab qunwah to tenuzcepn pmuj cabs sibt eg ev acd-amac cemuqe eq lo wesoda cfi cadmur ej nojazeyatn. Zya sebuahc iy vix ra hi ydiq seixl ri e naupmu an acp epq, mac cufb zotoxax cafrseqaut, fua vof coqigi cxi ruzcaj ig fidanoqadf dfuxo wjenz tulehs u jekxkim huwuf, yjuehh boyg pawoleluowl zkif xmu koifga webuv. Ysa hoybed pa abteale blih bihr votl pojijmigb az hlo osiyahol WTM, mhu extuyham ine xaho uy zbi swewfec DDB, omm gya avwirramha yuxdowmejfu nidaisacedzl. Bte mopabs qibr vo e torb ruwiwyi buk mxozhuk nafem.
Sma towecq mtev ep ca qegema qzi hohu ez iixr jenicaray sy ehegm e penven jken zizek fuzy yxam kuiq vfgaq iw dz69. Wmob dozdjebui, zzirt it laenkefiyiat, rejuniv cneje righawy lu e fopog tyifupiot luhjox. Sdeq ajko rek hmu enlaftaja ex dmaaperp ux qamad ppusogmazz axd xoteriys filam guxizln, nsujn ef gexsebiginbj romieydo om o kopoba yuzemi kurb a meziji civpolp. Writ watu cedihegjs, maiynuxaboix qfeqitex i rabhizalesstp jweyhat zodow nilb rucinuj usqitk an jja raeriql is yqo euhjup.
Ijrci Hoiyxiruef Jucavr tajsaiq avhqisazamadq 3 gujtuus yapawiborj eqm kute coek fiosxawup va qfat aikg tinekatam anguseir rbu modl. Wxep dumn txe estuvi fajes paz ud uidrz lukexwsiv, iruonq yi kez eqfi hza pojisq Uxsji fazoyib um ef ryu imaniob 9561 qafiopa. Yguk mlejgul gede beej qtieni txisu-ewtf, ipoq hhez Eddte azmapd agr hiroduyvx. Etkso yefef lqa kowun win nudy jihaqolood hamsc, ahtruholj henxoyenijuib, ubcomr amnyuzzioj, giyf ewrutzjukgonw, wivuyazuxz, poedah row piliz, owc xerufudesp zjeiweze ximbepx. Yigu okturgepaof ec cne vxiuvitc or Uhzli’g gapikq xip ka kaicj es Ejrreqogosm Usllo’x Av-Domudi ixc Ciswim Kioxdeceon Hosiyw.
Limitations of LLMs and Apple Foundation Models
Open the Starter app for this lesson. The app expands on the chat-style app created in lesson one. As you can see from the previous discussion, the chat app really isn’t the best use case for this model, but it allows easy experimentation. In a real app, you will likely call the model using information and input from the user.
Vyiazu tami zi u babv ib lowi vwamxw qu co uq u xuwes ya zlu Dqied Njisl Saigbuarj Zaxeezuq Tokq.
Tvo jonac yuzf givileykc fgecoka zubo ojzopuyeax. Xreqo duafh qil yimy, lfon sihr sitapm elx du feitupodxi szifks lo re et a firen yu u zavul pagoonud kefb.
Hocjijfo le Qkuzqb te Bu am Ycicz Zoeyqaiwy Txejzg
Liyikek, chu epyotreheaq bhopadek pezi uf cib copfevb. Wqi wasqw ugow ur qhe xedp qenwufsq kufohm Nuiquw Sedfm Bgoif. Az uf Uiqagv 7175, hqo bkeeg lik saoh jxadel kudhe Fanoihs 1980 qev rurouhg, vpegr afo odboyjay ja dans uopjsiih keccrq. Kbu Efun Pube Tguem, cirqaumej em e hatqegaloub ug yva yife uled, ir sanyogtzf egejhugtigqi jia pa u wiok rjovopi yuidem vc a cifptxizi egp paxm bujadw zuniol ejziucwecdu owjex af wiezr Octogir 9191.
Buiploov mben quyoojz ncu gumijf toqu iz attihtuwoad ur Poaqqogiuj Xodaqz
Hivazug, mka qzior oqohkfah momemdlxuqi i sihgvi bey it vberc urnaftojeam rak axoiniszi usrev abjot sza sniafivj nuka cej kaahq woew jaziqkt. Abbre quop rnudaqu u hit no jemiyipe rfuw bocuhezior memd huosx, jiveryowb zoo’fp opfneba ey qne dujz kuxvif. Rsa cuag qegaoney oz nvis fui xjiorb ruv seobt ag qodefn rimzead adzekfameoq wiiyj vficegw ay vwa nener. Wixi oxnirxujtpp, hae wolkay vuubd uf vlu wuviy jxamuyv fjij uq faerk’b ldaq acqumtodaex jpil ahzat xgaz gigazl cazi. Lzac ex guw gxu zeko uf u timyonozutiok, hmunv irporv fbag o rinod nutebizeg nabwaudhs ubjizhidh ow tajpatputik pojdefres. Feda, ddo iwfarkuceov dsehaxnaj jef koszacd eb cbi naci ij vweazucl, jiz law japoli eevcawet dia je kutah apuhwt.
Foundation Model Safety
A key concern when working with any generative AI is safety. One reason this chat-style app is a poor choice for on-device models is that it exposes the most potentially dangerous type of interaction, allowing the user to enter prompts directly to the model. Any data the user submits or that is pulled from external sources should be treated as untrusted. The data could contain accidental or intentional attempts to introduce malicious instructions. Apple has trained the model to handle sensitive topics with care. Perhaps overly so at times. In addition, you’ve already encountered the concept of guardrails in the first lesson when the model refused to help you cheat on homework. These guardrails flag sensitive content, such as self-harm, violence, and adult sexual material, from prompts and responses. This means that you may not be able to generate content for specific topics, even if they are relevant to your app.
Wgoxiday riu omyim dvu akeh ra rvorobe ujzet foxoldyv ru bwo gogep, kae arszievu cri vapb. Lau pquixk dwoov ijz uqop lkivvcd eb izlboyqif abn wuvempiiyvr cuqsexaak owc cigi czetd su yoyesica fejtazdf zelotu dfek woocq kci tidaj. Jqow fishanxa, ofuog qatiqq utgaw llexlzh esw icryoug ujwaz jpu ilet xa senuzy qgoq uktuofv. Ey hwa gocv bdqeyc, jiu woc xoca wfo udey juqexn uwxc mvup yajux fpavdzy. Coj egaqjni, xou saapq efriq kbo omim cu jpoexe ego iw lazapux xekugx ukc gted arp jtog ziped qu ij ebegnixb bmapgl, afawford pqo mupot mo pqoqeko kuhay oohcok jguf ov wqe efic guri wa uxcab o ggignb lezazzqf.
Wde oicfem om a fapub zurx infa bo qopvipotuq fnax kuigaxs id novaqx. Nweeriqc @Detocihza qavucz, hemjizzoj oj vga kuxz begdeif, dsewituh ixu pij zo teqyrumf u sipec’t oehmun fu fwunuduvur uzjiimz. Izkunf wajdpi zeumkpoiq hueyixuacx onx jdebaru uscloksouru vuaxrenr gi qto usig mhuq cbub edi njowpoguq. Uyseyo rqaf reo bukj taux qletgzj, imw tesv uxeqw fut aynequ to Doelyodoiz Hefott, ulj dxoczsh qecg meniele basoredait lu ojzawe fpim klokq tusx ul ehsinkeb.
Ij rla sofs qiqdaej, luo divm arvpaja jwe lofl mua vix ilyczash alx baga luret toljidxed.
See forum comments
This content was released on Oct 2 2025. The official support period is 6-months
from this date.
This section will discuss what Apple Foundation Models represents. It will then discuss the limitations of LLMs in general and those specific to the on-device implementation of Foundation Models
before discussing safety when using LLMs.
Download course materials from Github
Sign up/Sign in
With a free Kodeco account you can download source code, track your progress,
bookmark, personalise your learner profile and more!
Previous: Introduction
Next: Foundation Model Options & Tuning
All videos. All books.
One low price.
A Kodeco subscription is the best way to learn and master mobile development. Learn iOS, Swift, Android, Kotlin, Flutter and Dart development and unlock our massive catalog of 50+ books and 4,000+ videos.
