Inkqubo yokuFowuna i-API yasimahla yabasebenzi be-Cloudflare AI: Ubhaliso, uQwalaselo, kunye nokuthunyelwa kwiNyathelo elinye

Bendisebenza kwiprojekthi yam ngenye imini, kwaye ndifuna ukwakha...AIUmncedisi kufuneka abize iimodeli ezinkulu. Xa ndibuza ngexabiso, wow, umnxeba ngamnye oya kwi-GPT-4o ubiza iisenti ezimbalwa. Ukuba iprojekthi yethu ibingasebenza ngokwenene, iifizi ze-API zodwa beziza kuba ngamawaka aliqela ngenyanga.

Sazibuza ukuba ikho na indlela engabizi kakhulu.

Emva koko sadibana neCloudflare Workers AI. Sasisazi ukuba iCloudflare yayibandakanyeka kwisigqibo se-AI ngaphambili, kodwa asizange siyithathe nzulu loo nto, sicinga ukuba iinkampani ezinkulu ziya kuhlawulisa iimveliso zazo. Kwavela ukuba, wow, banikezela ngeeNeurons ezili-10000 simahla yonke imihla.

Ithetha ukuthini i-10000 Neurons? Ziingxoxo ezimalunga namakhulu ambalwa okanye iiseshoni zokuvelisa imifanekiso ezingamakhulu ambalwa ngosuku. Oko kwanele ukuba umntu omnye adlale ngayo.

Ndavuya kakhulu ndaza ndaqalisa ngoko nangoko ukuphanda ngayo.

Yintoni i-AI yabasebenzi beCloudflare?

I-AI yabasebenzi be-Cloudflare, ewe, ndicinga ukuba...Ukubekwa endaweniOko kunomdla kakhulu. Ayisiyo nje umboneleli wemodeli olula; liqonga le-AI inference elisebenza kwinethiwekhi ye-Cloudflare yomhlaba wonke. Nge-edge nodes kwizixeko ezingaphezu kwama-300, isondele kakhulu kubasebenzisi.

Cinga ngale nto: xa usebenzisa i-API ye-OpenAI, isicelo siya e-US kuqala, sicutshungulwe, size sibuye, sithathe ii-milliseconds ezingamakhulu amathathu ukuya kwamane. Kwelinye icala, i-Cloudflare ngumboneleli we-CDN onama-node kwihlabathi liphela. Naphi na apho ukhoyo, isebenzisa imodeli kwi-node ekufutshane.

Ixesha eliqhelekileyo lokuphendula lingaphantsi kwe-100 milliseconds. Lo mahluko uyabonakala kumsebenzisi.

Okumnandi ngayo kukuba isiqalo sayo esibandayo sikwinqanaba le-millisecond, kwaye inyuka ize ihle ngokuzenzekelayo. Nokuba kukho ithrafikhi engxamisekileyo, ayizukulibaziseka, ngokungafaniyo namanye amaqonga akhawuleza kakhulu xa kungekho mntu uyisebenzisayo ngo-3 AM, kodwa aphinde aqhekeke ngokupheleleyo ngexesha leeyure ezixakekileyo emini.

Iimodeli ezingaphezu kwama-50, ezigubungela zonke iimeko

Kukwakho neemodeli ezininzi ezixhaswayo; ukusukela ngoJuni 2026, sele zingaphezulu kwama-50.

Kwincoko esekelwe kumbhalo, i-Llama 3, i-Llama 4, i-Mistral, i-GLM, i-Qwen, i-Gemma, kunye ne-Deepseek-r1 zonke ziyafumaneka. Kwimveliso yomfanekiso, kukho i-Stable Diffusion, i-FLUX, kunye ne-Pixverse. Kwintetho, kukho i-Whisper kwintetho-ukuya-kumbhalo, i-TTS kwintetho-ukuya-kumbhalo, kwanokuveliswa kweevidiyo.

Ngaphezu koko, i-API yayo yenzelwe ukuba icoceke kakhulu. Nokuba yeyiphi na imodeli esetyenzisiweyo, isebenzisa ujongano oludibeneyo; udinga ukutshintsha igama lemodeli kuphela.

env.AI.run ("igama lomzekelo", {…})

Ukutshintsha imodeli ngomgca omnye wekhowudi kulula kakhulu.

Ixabiso eliphantsi kakhulu, umda wetyala wasimahla omkhulu

Ndiyenzile nexabiso. Isibonelelo sasimahla semihla ngemihla yi-10000 Neurons, nto leyo eyaneleyo kakhulu kuvavanyo lomntu. Ukuba ufuna ukuhlawula, yi-$0.011 kwi-1000 Neurons, nto leyo ebiza kancinci ngama-60% ukuya kuma-90% kune-OpenAI. Ukongeza, ihlawuliswa yi-Neurons, kungekhona ngamathokheni, nto leyo eyenza ukuba ingabizi kakhulu kwiincoko ezincinci.

Inene, ndicinga ukuba le ndlela yokuhlawula ilungile. Inani leeNeuron ozisebenzisayo liyahluka ngokuxhomekeke ekubeni unencoko elula okanye ende kakhulu, kodwa umahluko awukho mkhulu njengakwi-token-based billing.

Okwangoku, usenokuba uzibuza, uyisebenzisa njani kanye kanye le nto?

Ndiyihambile le nkqubo, mandikuxelele ngayo.

Bhalisa kwaye wenze iithokheni ze-API

Inyathelo lokuqala kukubhalisa iakhawunti yeCloudflare. Ukuba sele unayo iakhawunti, ngena nje. Ukuba akunjalo, ukubhalisa kulula; gcwalisa nje idilesi ye-imeyile uze usete igama lokugqitha.

Emva kokungena, cofa i-AI kwimenyu esekhohlo, uze ucofe i-Workers AI.

Nje ukuba ungene ngaphakathi, uza kubona iqhosha lokwenza ithokheni ye-API. Cofa kulo ukuze wenze ithokheni.

Nantsi inkcukacha ekufuneka uyiqaphele: ithokheni evelisiweyo inokubonwa kube kanye kuphela ngexesha lokudalwa, ngoko ke qiniseka ukuba uyayigcina. Eli phepha liza kukunika ne-ID yeakhawunti; uza kuzifuna zombini ezi zinto kamva.

Emva kokugcina, uza kubona umzekelo wokusetyenziswa ezantsi kwephepha, ongumyalelo we-curl. Yitshintshe nge-ID yeakhawunti yakho kunye ne-APIKEY. Lo myalelo ujongeka ngolu hlobo:

curl \
  https://api.cloudflare.com/client/v4/accounts/账户ID/ai/run/@cf/模型ID \
  -H "Authorization: Bearer 刚才生成的APIKEY" \
  -d '{"messages":[{"role":"system","content":"You are a friendly assistant that helps write stories"},{"role":"user","content":"Write a short story about a llama that goes on a journey to find an orange cloud"}]}'

Jonga uluhlu lweemodeli ezikhoyo

Emva koko kuphakama umbuzo: uyifumana phi i-ID yomzekelo?

Kukho iqhosha loxwebhu kwicala lasekunene lephepha le-Workers AI.

Cofa kudidi "lweeModels"; kulapho uza kufumana khona uluhlu lwazo zonke iimodeli.

Khetha leyo ufuna ukuyisebenzisa, kwaye ungayibona i-ID yomzekelo ngokucofa kuyo.

Vavanya ukubiza imodeli yeKimi 2.6

我选择用kimi2.6来测试一下。把模型ID复制出来，替换到curl命令里，从@cf/开始替换。

Isicelo sathunyelwa, kwaye isiphumo sabuyiselwa ngokukhawuleza.

Isantya sempendulo sikhawuleza kakhulu, sithambile kunokuba bendilindele.

Inyaniso kukuba, bendifuna nje ukubona ukuba le khredithi yasimahla ingenza ntoni, kwaye kwavela ukuba inamandla kakhulu. Yanele ngokupheleleyo incoko yemihla ngemihla, ukuvelisa umxholo, ukuguqulela, kunye nokubhala iikhowudi.

Kwaye cinga ngayo: le nto isebenza kwiindawo ezisemaphethelweni. Usebenzisa usetyenziso lwe-AI kubasebenzisi kulo lonke ilizwe okanye kwihlabathi liphela, kwaye isantya sempendulo sihlala sihambelana, ngaphandle kokukhathazeka ngemiba yokuthunyelwa kwengingqi. Oku kunomtsalane kakhulu kubaphuhlisi ngabanye kunye namaqela amancinci.

Ngaphambili, ukuba ufuna ukusebenzisa imodeli, kwakufuneka uthenge iGPU uze wakhe eyakho, eyayibiza kakhulu, okanye usebenzise i-GPU instance evela kumboneleli welifu, eyayinzima ukuyicwangcisa. Kwelinye icala, iCloudflare ikulungele konke. Ubhala nje ikhowudi ukuyibiza, kwaye iphatha zonke izinto ezimdaka kuwe.

Ndicinga ukuba le ndlela yokucinga ichanekile. Amandla e-AI aza kufana neziseko ezifana namanzi nombane; akufuneki wazi ukuba umbane wenziwa njani, ungawufaka nje uwusebenzise. I-Cloudflare Workers I-AI ithatha le ngcamango "yokufaka udlale" iye kwinqanaba eligqithisileyo nangakumbi.

Kakade ke, ayinazo iingxaki zayo. Isabelo sasimahla yi-10000 Neurons kuphela, ngoko ke kuya kufuneka uhlawule ukuba ufuna ukuyisasaza kwindawo yemveliso enetrafikhi ephezulu. Kwakhona, nangona kukho iimodeli ezingaphezu kwama-50, xa kuthelekiswa ne-ecosystem ebanzi ye-OpenAI, ezinye iimodeli ezincinci okanye ezintsha azixhaswa.

Kodwa kwiiprojekthi zobuqu, izixhobo ezincinci, okanye iimvavanyo eziqhutywa ngumdla, ndicinga ukuba kwanele.

Yiloo nto ndiyithethayo ngokunyaniseka. Ekubeni sele uyifundile le ncwadi, ukuba uyifumene iluncedo, nceda uyithande kwaye uyibelane ngayo. Ukuba ufuna ukufumana uhlaziyo kuqala, ungandilandela! ⭐

Enkosi ngokufunda inqaku lam. Siza kubonana kwixesha elizayo.