Uyifaka njani iGemma 3 LLM Windows 11 inyathelo ngenyathelo

Imifuziselo yolwimi lwemithombo evulelekileyo iye yavela ngokutsiba nemida, kunye Namhlanje kunokwenzeka ukuzonwabela ngokuthe ngqo kwikhompyuter yakho ngaphandle kokuxhomekeka kwiinkonzo zamafu.. Enye yezona zinto zithembisayo ngoku yiGemma 3, iLLM entsha kaGoogle esekwe kwitekhnoloji yeGemini, ebalaseleyo ngokukwazi kwayo ukuseta umbhalo kunye nemifanekiso, kunye nefestile enkulu yomxholo ukuya kuthi ga kwi-128k iithokheni kwiinguqulelo zayo eziphambili. Ngolwazi oluthe kratya malunga noku kukhululwa, ungandwendwela inqaku lethu umboniso weGemma 3.

Ukuba uyasebenzisa Windows 11 kwaye ujonge ukufaka iGemma 3 yovavanyo okanye imveliso yasekhaya, ufike endaweni elungileyo. Makhe siye ngaphaya ngokweenkcukacha zonke iindlela ezinokubakho zokuyinyusa kwaye isebenze kwikhompyuter yakho, kubandakanya ezona khetho zicetyiswayo njengeOllama, iLM Studio, kunye nenye indlela esekwe kwilifu ngeGoogle AI Studio. Ukongeza, siya kujongana neemfuno zobugcisa, iingenelo zendlela nganye kunye Indlela yokwenza uninzi lwesakhono sobukrelekrele bokwenziwa bunamandla.

Yintoni iGemma 3 kwaye kutheni uyifake?

IGemma 3 sisizukulwana sesithathu seemodeli zeLLM ezikhutshwe nguGoogle phantsi kwelayisensi evulelekileyo.. Ngokungafaniyo nezisombululo zangaphambili ezifana neLlama okanye iMistral, inika inkxaso ngokuthe ngqo kwigalelo lomfanekiso, umxholo obanzi kakhulu, kunye nenkxaso yeelwimi ezingaphezu kwe-140. Ngokuxhomekeke kwiimfuno zakho, unokukhetha kwiinguqulelo ezininzi zemodeli ukusuka kwi-1B ukuya kwi-27B iiparamitha:

• Gemma 3:1B: Imodeli ekhaphukhaphu ilungele imisebenzi esisiseko kunye neendawo ezinobuncwane obulinganiselweyo.
• Gemma 3:4B: Ukulinganisa ukusebenza kunye nokusebenza kakuhle kwizicelo eziphakathi.
• Gemma 3:12B: Icetyiselwe uhlalutyo oluntsonkothileyo, ukucwangciswa kwenkqubo kunye nokusetyenzwa kweelwimi ezininzi.
• Gemma 3:27B: Olona khetho lunamandla, lwenzelwe ukusetyenziswa okunzulu, ukusetyenziswa kwe-multimodal kunye nomthamo omkhulu womxholo.

Ukuba nokwenzeka kokuqhuba imodeli ephambili enjalo kwiPC yakho Itshintsha imithetho yomdlalo ngokwemigaqo yabucala, isantya sokuphendula, kunye nokuxhomekeka kumaqela esithathu. Akusekho mfuneko yokuba uhlawule imirhumo yenyanga okanye uyeke idatha yakho. Ekuphela kwento efunekayo kukulungiselela okuncinci kunye nomnqweno wokufunda. Ukuba ufuna ukwazi ngakumbi malunga neenzuzo zezi modeli, jonga inqaku lethu vula iimodeli ze-AI ezinobunzima.

Inketho 1: Ufakelo kunye no-Ollama

I-Ollama mhlawumbi yeyona ndlela ilula yokuqhuba iLLM njengeGemma 3 ukusuka Windows 11. Ujongano olusekwe kwisiphelo sendlela ikuvumela ukuba ufake kwaye uqhube imifuziselo ngomgca womyalelo olula. Ukongeza, iyahambelana ne-macOS, iLinux, kunye neWindows, iyenza kube lula ukuyisebenzisa kwiindawo ezahlukeneyo.

Amanyathelo okufaka uOllama kwaye usebenzise iGemma 3:

1. Ngena kwiwebhusayithi esemthethweni: ollama.com.
2. Khuphela isifakeli seWindows kwaye uyiqhube njengayo nayiphi na enye inkqubo.
3. Vula i-Command Prompt (CMD) okanye i-PowerShell kwaye uqinisekise ukufakela nge:

ollama --version

Ukuba konke kuhamba kakuhle, ngoku unokukhuphela naziphi na iitemplate zeGemma 3 ezikhoyo. Sebenzisa ngokulula omnye wale miyalelo ngokuxhomekeke kwitemplate oyifunayo:

ollama run gemma3:1b
ollama run gemma3:4b
ollama run gemma3:12b
ollama run gemma3:27b

Nje ukuba ukhutshelwe, unokuqalisa ngokulula imodeli.. Ukwenza oku, sebenzisa:

ollama init gemma3

Ukusukela ngaloo mzuzu ukuya phambili, unokuqalisa ukusebenzisana neLLM ngo:

ollama query gemma3 "¿Cuál es la capital de Japón?"

Ukuba ufuna ukuthatha ithuba lemisebenzi ye-multimodal, ungasebenzisa imifanekiso kwimibuzo yakho:

ollama query gemma3 --image "ruta-de-la-imagen.jpg"

Yintoni oyifunayo ukuze usebenze kakuhle? Nangona u-Ollama enganyanzelisi ubuncinci beemfuno ezingqongqo, iimodeli ezinkulu (ezifana ne-27B) zifuna ubuncinci be-32GB ye-RAM. Nge-16GB unokusebenza ngaphandle kweengxaki ngemodeli ye-7B, kwaye nangona ukusetyenziswa kwe-GPU kunganyanzeliswanga, kunceda kakhulu ngesantya.

Indlela yesi-2: Sebenzisa i-LM Studio

I-LM Studio sesinye isixhobo sasimahla esikuvumela ukuba ufake kwaye uqhube imifuziselo yeLLM ekuhlaleni ukusuka kujongano lomzobo.. Iyahambelana neWindows, macOS, kunye neLinux, kwaye inzuzo yayo enkulu kukuba ayifuni ulwazi lobuchwephesha ukusebenza.

Imiyalelo:

1. Khuphela i-LM Studio kwiwebhusayithi yayo esemthethweni: lmstudio.ai.
2. Faka kwaye uyiqhube.
3. Cofa kwi icon yeglasi ethi "Fumana."
4. Chwetheza "Gemma 3" kwi-injini yokukhangela ukuze ubone iimodeli ezikhoyo.

Ngaphambi kokufaka, khangela ukuba imodeli iyahambelana nesixhobo sakho. Ukuba ubona isilumkiso "Mhlawumbi mkhulu kakhulu kulo matshini," usengawufaka, kodwa ukusebenza kakuhle akuqinisekwanga.

Nje ukuba imodeli ehambelanayo ikhutshelwe:

• Cinezela "Layisha iModeli" ukuyilayisha.
• Okanye vula incoko entsha kwaye ukhethe imodeli kwimenyu eyehlayo.

Eyona nto ilungileyo nge-LM Studio kukuba isebenza njengohlobo lweChatGPT yalapha, ngaphandle kweintanethi nangolwimi lwakho.. Unokwenza iincoko ezininzi kwaye ugcine iincoko zakho ukuba uyafuna. Ukongeza, ukuba uvumela ukhetho lwe "Local Server", unokuludibanisa kunye nezicelo zakho zePython usebenzisa i-API ehambelana ne-OpenAI.

Ukhetho 3: Sebenzisa iGoogle AI Studio (kwi-intanethi)

Ukuba awukwazi okanye awufuni ukufaka nantoni na, ungasebenzisa iGemma 3 ngqo kwilifu ngeGoogle AI Studio.. Akukho kufakelo olufunekayo, kodwa uqhagamshelo lwe-Intanethi kunye neakhawunti kaGoogle ziyafuneka.

Kufuneka uye nje aistudio.google.com kwaye ukhethe "Gemma 3" kuluhlu lweemodeli. Ukusukela kuloo mzuzu ukuya phambili, ungaqala ukuncokola ngemodeli ngokungathi luguqulelo oluphambili lweBard okanye iChatGPT, kubandakanya nokufakwa komfanekiso.

Ufakelo lwamafu ngeNodeShift (ukhetho)

Kwabo bafuna amandla amaninzi okanye ukuhambisa imodeli ngobungcali, kukho ukhetho lokusebenzisa iinkonzo zelifu njengeNodeShift. Ngabo, unokuqesha oomatshini abaneGPU ezinamandla kwaye uqwalasele indawo yakho efanelekileyo ukuze usebenzise iGemma 3 ngaphandle kwemida.

Amanyathelo asisiseko kwi-NodeShift:

1. Yenza iakhawunti kwi app.nodeshift.com.
2. Qala i-GPU Node yesiko (umzekelo nge-2x RTX 4090).
3. Khetha umfanekiso osele uqwalaselwe ngo-Ubuntu + Nvidia CUDA okanye iJupyter Notebook, kuxhomekeke ekubeni uya kusebenzisa iOllama okanye iTransformers.
4. Qhagamshela nge-SSH kwaye ufake imodeli ukusuka kumgca womyalelo.

Olu hlobo lofakelo lukunika ukufikelela kuqwalaselo lobuchwephesha, ilungele imifuziselo yoqeqesho, ukuvavanya ukusebenza, njl.njl. Nangona ingeyomfuneko kubasebenzisi basekhaya, iluncedo kwabo bafuna ukulinga nzulu okanye ukwakha usetyenziso kwii-LLM eziphambili.

Iimfuno zenkqubo kunye neengcebiso zobugcisa

Ayizizo zonke iimodeli zeGemma 3 eziya kusebenza kuyo nayiphi na iPC. Ngezantsi sikushiyela ireferensi ngokubanzi ngokohlobo lwemodeli:

• ukuba iimodeli 1B ukuya 7B: ubuncinane 8 GB ye RAM. Basebenza phantse kuyo nayiphi na iPC yanamhlanje, nangaphandle kweGPU.
• ukuba Iimodeli ze-13B: ziyacetyiswa I-16GB ukuya kwi-24GB RAM.
• ukuba Iimodeli ze-27B: ziyafuneka ubuncinci i-32 GB ye-RAM kwaye kukhethwa i-GPU ezinikeleyo.

Ukuba ne-RAM eninzi kukhawuleza ukusebenza kwaye kuthintela iimpazamo ngenxa yokunqongophala kwememori. Ngelixa u-Ollama kunye ne-LM Studio bezama ukusebenzisa izixhobo ngokufanelekileyo, kuya kuxhomekeka kakhulu kwi-hardware yakho. Ukongeza, isantya sokuphendula siphucula kakhulu ukuba iGPU isetyenziswa endaweni yeCPU.

Ukufaka iGemma 3 Windows 11 kulula kunokuba kubonakala.. Ayinamsebenzi nokuba ugqibe ekubeni usebenzise i-Ollama ngokulula kwayo, i-LM Studio kujongano lwegraphical, okanye iGoogle AI Studio ukuyidlala ngokukhuselekileyo efini. Into ebalulekileyo kukuba indlela nganye ilungelelanisa kumanqanaba ahlukeneyo amava kunye nobuchule bobugcisa. Ngoku ukuba uyazazi zonke iinketho kunye nento oyifunayo ukuze uqalise, ungaqala ukuzama obu bukrelekrele bokwenziwa basekhaya namhlanje.