TurboLLM: Offline AI Chat

I-TurboLLM isebenzisa amamodeli ezilimi ezinkulu ngqo ku-CPU yefoni yakho ye-Android — akukho ifu, akukho akhawunti, akukho ukubhalisa, futhi akukho uxhumano lwe-inthanethi oludingekayo uma imodeli ilandiwe.

KUNGANI I-AI YASENDAWENI

Izinhlelo zokusebenza eziningi ze-"AI chat" ziyi-wrapper encane ezungeze i-API yamafu yomunye umuntu: izixwayiso zakho ziyaphuma efonini yakho, izingxoxo zakho zihlala kuseva ongayilawuli, futhi uhlelo lokusebenza luyeka ukusebenza ngokushesha lapho ulahlekelwa isignali. I-TurboLLM ihlukile. Imodeli isebenza enjinini yangempela yokuphetha kudivayisi (llama.cpp, ehlanganiswe i-Android), futhi ingxoxo yakho akudingeki iphume efonini yakho.

LOKHO ONGAKWENZA

- Xoxa nama-LLM endawo avulekile — i-Llama, i-Qwen, i-Gemma, nanoma iyiphi enye imodeli ye-GGUF — ungaxhunyiwe ngokuphelele, ngisho nakumodi yendiza.

- Thola futhi ulande amamodeli ngqo kusuka ku-Hugging Face, ngaphakathi kohlelo lokusebenza, ngokuthepha okukodwa.

- Injini yokuphetha izifaka ngokuzenzakalelayo okokuqala uma uyidinga — akukho ukusethwa ngesandla, akukho ukuhlanganiswa, akukho ukuphela.

- Dala ama-ejenti e-AI ngokwezifiso ngokufinyelela kwawo kwesistimu kanye nekhono/amathuluzi, emisebenzini ehlukene - usizo lokubhala ikhodi, ukubhala, ucwaningo, noma i-slate engenalutho ngokuphelele ngaphandle kwemiyalelo efakiwe.

- Landelela ukusetshenziswa kwakho ngokuhamba kwesikhathi: amaseshini, imiyalezo, amathokheni aphelele akhiqizwe, imigqa, kanye nemodeli yakho esetshenziswa kakhulu - konke kubalwa endaweni kusuka emlandweni wakho.

- Lungisa ukukhiqizwa ngezilawuli zengxoxo ngayinye: isabelomali sokucabanga, ukusampula, kanye nomzamo wokucabanga, kumamodeli awasekelayo.
- I-interface efanayo nohlelo lokusebenza lwedeskithophu lwe-TurboLLM (i-Windows, i-Mac, i-Linux) - uma usebenzise i-TurboLLM kukhompyutha, ifoni yakho isebenza ngendlela efanayo.

UBUMFIHLO NGOKWENZA

- Akukho akhawunti, akukho ukungena ngemvume, akukho backend yefu yezingxoxo zakho.
- Akukho telemetry, akukho analytics, akukho SDK yokubika ukuphahlazeka eshayela ekhaya kulesi sakhiwo.
- Izingxoxo namamodeli alandiwe ahlala kuphela kwisitoreji sangasese salolu hlelo lokusebenza kudivayisi yakho.
- Imiyalelo yakho, amafayela, kanye nesisindo samamodeli akukaze kuthunyelwe kithi noma kunoma ubani omunye. Izicelo zenethiwekhi kuphela ezenziwa yilolu hlelo lokusebenza yilezo oziqalile wena: ukulanda imodeli oyikhethile, ukulanda i-binary yenjini yokukhomba okokuqala uma kudingeka, noma ithuluzi elikubiza ngokuthi uvunyelwe ngokusobala.

OKUFANELE UKWAZI

I-Inference isebenza ku-CPU yefoni yakho, ngakho-ke isivinini sincike ku-chipset yakho kanye nosayizi wemodeli oyikhethayo - amamodeli amancane (amapharamitha angu-1-4B) azizwa esabela kakhulu kumafoni amaningi, kanti amakhulu akhona uma ihadiwe yakho ingakwazi ukuwathwala. Ukukhiqiza okusebenzayo kuwumsebenzi wangempela wokubala, ngakho-ke lindela ibhethri elijwayelekile kanye nomthwalo wokushisa, njenganoma yimuphi umsebenzi we-AI kudivayisi.

KUNGOKUBANI

Abasebenzisi abaqaphela ubumfihlo, abathanda i-AI yendawo, abathuthukisi asebevele besebenzisa i-TurboLLM noma i-llama.cpp kukhompyutha, nanoma ubani ozibuza ukuthi ingxoxo ye-LLM engaxhunyiwe ku-inthanethi izwakala kanjani kufoni - ngaphansi kolawulo lwakho, kuhadiwe yakho, ngaphandle kokuphuma kudivayisi.