Ollama AI: Ubukrelekrele bokwenziwa kwitheminali

I-Ollama ikuvumela ukuba usebenzise iimodeli zeArtificial Intelligence ekuhlaleni


Ukusukela ekuveleni kweChatGPT, kuye kwavela izixhobo ezininzi ezihlonipha ubumfihlo babasebenzisi. Ngethamsanqa, iiprojekthi ezininzi zomthombo ovulekileyo zavela ukulungisa oku. I-Ollama AI sisixhobo sokusebenzisa ubukrelekrele bokwenziwa kwi-terminal

Insiza iya kuthi ikuvumela ukuba usebenzise uluhlu lweeLLMs kwitheminali yekhompyuter yakho okanye isixhobo sebhodi enye nje ukuba ihardware yakho inamandla ngokwaneleyo. Into entle kukuba ilayibrari yabo ineenketho zabo bonke ubukhulu.

Ii-LLM (IiModeli zoLwimi oluKhulu) zisetyenziselwa ukuqonda ukuvelisa nokuguqulela ulwimi kwinqanaba elikufutshane nolwimi lwendalo. Baqeqeshwe ngezixa ezikhulu zedatha yokubhaliweyo. Le mifuziselo ingasetyenziselwa imisebenzi ebanzi eyayamene nolwimi, efana nokuphendula imibuzo, ukushwankathela isicatshulwa, ukuguqulela kwezinye iilwimi, ukuvelisa umxholo ohambelanayo noyilayo, ukugqibezela izivakalisi, kunye nokufumana iipateni zolwimi.

Intetho ethi "Isikali esikhulu" ibhekisa kuzo zombini isixa esikhulu sedatha (iibhiliyoni) kunye neeparamitha ezisetyenziswa ngumfuziselo ukuseka ubudlelwane kunye nokufumana iipateni phakathi kwedatha.

Iparameters yinto ekufuneka ithathelwe ingqalelo. Nangona kukho abo baxela ukuba basebenzise i-Ollama AI kwiRaspberry Pi 4, ayizizo zonke iimodeli ezinokusetyenziswa ukuba awunamemori yaneleyo. Ndisebenzise iimodeli ezine-7 yeebhiliyoni zeeparamitha ezine-6 ​​GB ye-RAM kwaye ngaphandle kwekhadi lemizobo ngaphandle kokuchaphazela ukusebenza kwenkqubo, kodwa ayikwazanga ukuphatha enye nge-13.
Iimodeli ezineeparamitha ezimbalwa zithetha iSpanish kwinqanaba labakhenkethi baseYankee kwaye azichanekanga kakhulu kwiimpendulo zabo, ke kufuneka ulumke xa uzisebenzisa.

Ollama AI: Sebenzisa ubukrelekrele bokwenziwa kwitheminali

Ngaphambi kokuba siqhubeke, masiqale ngengcaciso emfutshane. I-LlaMa yi-algorithm yokufunda ngomatshini evumela i-Artificial Intelligence ukuba ifunde ukuqaphela nokuhlela imifanekiso yezinto.

Kolu hlobo lweemodeli Baqeqeshwa ngemifanekiso ephawulweyo ukuze bafunde ukuqaphela iimpawu ezithile ezifana nokuthungwa, imilo kunye nombala ovumela ukuba bachongwe kwezinye iimeko.

Ekubeni igama le-algorithm libizwa ngokufanayo negama lesilwanyana, ezinye iilayibrari zathiywa ngeentlobo ezifanayo ezifana ne-vicuna (Vicuña kwiilwimi ezingasebenzisi i-ñ) okanye i-Alpaca.

Ukubuyela kwi-Ollama AI, Ivumela ukuba sikhuphele kwaye siqhube iimodeli ezahlukeneyo zemithombo evulekileyo ukusuka kwi-terminal yokuhanjiswa kweLinux. Ifakwe ngomyalelo:

curl https://ollama.ai/install.sh | sh

Ukuqalisa imodeli sibhala umyalelo:
ollama run nombre_del_modelo

Sinokubona uluhlu lweemodeli Apha

Umzekelo, ukufaka iTinyLlaama, imodeli engaphezulu nje kweterabyte yeeparamitha, sisebenzisa lo myalelo:

ollama run tinyllama

Ngexesha lokuqala lo myalelo uphunyezwa imodeli ikhutshelwa kwikhompyuter. Nceda uqaphele ukuba ezinye zithatha iigigabytes ezininzi zediski yendawo.

Ungayicima imodeli ngomyalelo:
ollama rm nombre_del modelo

Ukubona uluhlu lweemodeli ezifakiweyo sebenzisa lo myalelo:
uluhlu lwe-ollama

Ungalubona ulwazi malunga nemodeli nganye efakiweyo nge:
ollama show

Inketho enomdla evela kwilayibrari yemodeli ye-Ollama yinto ebizwa ngokuba yi "Uncensored".. Iimodeli ezaziwa kakhulu zeArtificial Intelligence zibeka izithintelo kwiintlobo ezithile zemibuzo ukuba ingahambelani nomthetho kuphela kodwa noxinzelelo lwezentlalo lokuchaneka kwezopolitiko.

Ndikhumbula ukuba kwakudala ndacela uCopilot, umncedisi weMicrosoft's ChatGPT-based, ukuba andixelele iziqhulo ze-iceberg.Wandixelela ukuba iintlekele zendalo zingumxholo onzima kakhulu ukuba wenze iziqhulo ngawo.

Iimodeli ezingavumelekanga Bachonga amaxesha apho ii-AI zenqaba ukuphendula okanye ukunika iimpendulo ezicalu-calulo kwiimodeli ezisisiseko, zisuse, kwaye ziqeqeshe inkqubo ukuba iphendule ngokuchanekileyo.

Kuya kufuneka ndenze uvavanyo olungakumbi ukuze ndazi ukuba i-Ollama AI inokuba luncedo kangakanani kwiikhompyuter ezinexabiso elincinci le-RAM kwaye ngaphandle kwekhadi lemizobo elizinikeleyo. Nakweyiphi na imeko, kulungile ukwazi ukuba iindlela ezizezinye zomthombo ovulekileyo zenza indlela yazo kwaye zivumela ukubalwa ukuba kupheliswe.


Yiba ngowokuqala ukuphawula

Shiya uluvo lwakho

Idilesi yakho ye email aziyi kupapashwa. ezidingekayo ziphawulwe *

*

*

  1. Uxanduva lwedatha: UMiguel Ángel Gatón
  2. Injongo yedatha: Ulawulo lwe-SPAM, ulawulo lwezimvo.
  3. Umthetho: Imvume yakho
  4. Unxibelelwano lwedatha: Idatha ayizukuhanjiswa kubantu besithathu ngaphandle koxanduva lomthetho.
  5. Ukugcinwa kweenkcukacha
  6. Amalungelo: Ngalo naliphi na ixesha unganciphisa, uphinde uphinde ucime ulwazi lwakho.