Inqubo Yokubiza I-API Yamahhala Yabasebenzi be-Cloudflare AI: Ukubhalisa, Ukucushwa, kanye Nokusetshenziswa Ngesinyathelo Esisodwa

Bengisebenza kuphrojekthi yami ngolunye usuku, futhi ngangifuna ukwakha...AIUmsizi udinga ukushayela amamodeli amakhulu. Lapho ngibuza ngentengo, wow, ucingo ngalunye oluya ku-GPT-4o lubiza amasenti ambalwa. Uma iphrojekthi yethu ibingasebenza ngempela, izimali ze-API zodwa bezizoba yizinkulungwane eziningana ngenyanga.

Sazibuza ukuthi ikhona yini indlela eshibhile.

Sabe sesihlangana ne-Cloudflare Workers AI. Sasazi ukuthi i-Cloudflare yayihilelekile ekucabangeni nge-AI ngaphambili, kodwa asizange sikuthathe ngokungathi sína, sicabanga ukuthi izinkampani ezinkulu zizokhokhisa ngemikhiqizo yazo. Kuyavela ukuthi, wow, banikeza ama-Neuron ayi-10000 mahhala nsuku zonke.

Kusho ukuthini ama-Neuron ayi-10000? Cishe kuyizingxoxo ezingamakhulu ambalwa noma izikhathi ezimbalwa zokukhiqiza izithombe ngosuku. Lokho kwanele ukuba umuntu oyedwa adlale ngakho.

Ngajabula kakhulu futhi ngaqala ukuyicwaninga ngokushesha.

Kuyini i-Cloudflare Workers AI?

I-AI Yabasebenzi Base-Cloudflare, ngicabanga ukuthi...UkubekaLokho kuyathakazelisa impela. Akuyona nje umhlinzeki wemodeli olula; kuyipulatifomu yokuqagela ye-AI esebenza kunethiwekhi yomhlaba wonke ye-Cloudflare. Njengoba inama-edge node emadolobheni angaphezu kuka-300, isondelene kakhulu nabasebenzisi.

Cabanga ngakho: uma usebenzisa i-API ye-OpenAI, isicelo siya e-US kuqala, sicutshungulwe, bese sibuya, sithathe ama-millisecond amathathu kuya kwamakhulu amane. I-Cloudflare, ngakolunye uhlangothi, ingumhlinzeki we-CDN onama-node emhlabeni wonke. Noma ngabe ukuphi, isebenzisa imodeli ku-node eseduze.

Isikhathi esivamile sokuphendula singaphansi kwama-millisecond ayi-100. Lo mehluko uyabonakala kumsebenzisi.

Okuhle kakhulu ngayo ukuthi ukuqala kwayo okubandayo kusezingeni le-millisecond, futhi ikhuphuka futhi yehla ngokuzenzakalelayo. Ngisho noma ithrafikhi ikhuphuka ngokuzumayo, ngeke ihlale isikhathi eside, ngokungafani namanye amapulatifomu ashesha kakhulu lapho kungekho muntu oyisebenzisayo ngo-3 ekuseni, kodwa bese ephahlazeka ngokuphelele ngezikhathi zomsebenzi omatasa emini.

Amamodeli angaphezu kuka-50, ahlanganisa zonke izimo

Kukhona futhi amamodeli amaningi asekelwayo; kusukela ngoJuni 2026, kakade asengaphezu kuka-50.

Ngengxoxo esekelwe kumbhalo, i-Llama 3, i-Llama 4, i-Mistral, i-GLM, i-Qwen, i-Gemma, kanye ne-Deepseek-r1 zonke ziyatholakala. Ukuze kukhiqizwe izithombe, kukhona i-Stable Diffusion, i-FLUX, kanye ne-Pixverse. Ukuze kukhulunywe, kukhona i-Whisper ukuze kukhulunywe kube umbhalo, i-TTS ukuze kukhulunywe kube umbhalo, ngisho nokukhiqizwe kwamavidiyo.

Ngaphezu kwalokho, i-API yayo yenzelwe ukuba ihlanzeke kakhulu. Kungakhathaliseki ukuthi imodeli esetshenzisiwe injani, isebenzisa isikhombikubona esihlanganisiwe; udinga nje ukushintsha igama lemodeli.

env.AI.run ("igama lemodeli", {…})

Ukushintsha imodeli ngomugqa owodwa wekhodi kulula kakhulu.

Intengo eshibhile kakhulu, umkhawulo wesikweletu wamahhala omkhulu

Ngiphinde ngenza intengo. Imali yamahhala yansuku zonke ingu-10000 Neurons, okwanele kakhulu ekuzihloleni komuntu siqu. Uma ufuna ukukhokha, kungu-$0.011 ngama-Neurons ayi-1000, okushibhile ngo-60% kuya ku-90% kune-OpenAI. Ngaphezu kwalokho, ikhokhiswa yi-Neurons, hhayi ngamathokheni, okwenza kube ngcono kakhulu ezingxoxweni ezincane.

Ngokweqiniso, ngicabanga ukuthi le ndlela yokukhokha ilungile impela. Inani lama-Neuron owasebenzisayo liyahlukahluka kuye ngokuthi uxoxa nje noma isikhathi eside kakhulu, kodwa umehluko awubi kakhulu njengasekukhokhisweni okusekelwe kumathokheni.

Kulokhu, ungase uzibuze, uyisebenzisa kanjani ngempela?

Ngidlule kule nqubo mathupha, ake ngikutshele ngayo.

Bhalisa futhi udale amathokheni e-API

Isinyathelo sokuqala ukubhalisa i-akhawunti ye-Cloudflare. Uma usuvele unayo i-akhawunti, vele ungene ngemvume. Uma kungenjalo, ukubhalisa kulula; vele ugcwalise ikheli le-imeyili bese usetha iphasiwedi.

Ngemva kokungena ngemvume, chofoza i-AI kumenyu yesobunxele, bese uchofoza i-Workers AI.

Uma usungaphakathi, uzobona inkinobho yokudala ithokheni ye-API. Chofoza kuyo ukuze udale ithokheni.

Nansi imininingwane okufanele uyiqaphele: ithokheni ekhiqiziwe ingabonwa kanye kuphela ngesikhathi sokudalwa, ngakho qiniseka ukuthi uyayilondoloza. Ikhasi lizokunikeza ne-ID ye-akhawunti; uzodinga kokubili lokhu kamuva.

Ngemva kokulondoloza, uzobona isibonelo sokusetshenziswa phansi kwekhasi, okungumyalo we-curl. Vele uwushintshe nge-ID yakho ye-akhawunti kanye ne-APIKEY. Umyalo ubukeka kanje:

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"}]}'

Buka uhlu lwamamodeli atholakalayo

Bese kuphakama umbuzo: uyitholaphi i-ID yemodeli?

Kukhona inkinobho yedokhumenti ngakwesokudla kwekhasi le-Workers AI.

Chofoza esigabeni esithi "Amamodeli"; yilapho uzothola khona uhlu lwawo wonke amamodeli.

Khetha leyo ofuna ukuyisebenzisa, futhi ungabona i-ID yemodeli ngokuchofoza kuyo.

Hlola 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, futhi umphumela wabuyiselwa ngokushesha.

Isivinini sokuphendula sishesha kakhulu, sibushelelezi kunalokho ebengikulindele.

Uma ngikhuluma iqiniso, bengifuna nje ukubona ukuthi le khredithi yamahhala ingenzani, futhi kuvela ukuthi inamandla kakhulu. Yanele ngokuphelele engxoxweni yansuku zonke, ekudalweni kokuqukethwe, ekuhumusheni nasekubhaleni amakhodi.

Futhi cabanga ngakho: le nto isebenza kuma-node asemaphethelweni. Usebenzisa uhlelo lokusebenza lwe-AI kubasebenzisi kulo lonke izwe noma ngisho nasemhlabeni jikelele, futhi isivinini sempendulo sihlala sinjalo, ngaphandle kokukhathazeka ngezinkinga zokuthunyelwa kwesifunda. Lokhu kukhanga kakhulu onjiniyela ngabanye kanye namaqembu amancane.

Ngaphambilini, uma ufuna ukusebenzisa imodeli, kwakufanele uthenge i-GPU bese uzakhela eyakho, eyayibiza kakhulu, noma usebenzise i-GPU instance evela kumhlinzeki wefu, okwakunzima ukuyilungisa. I-Cloudflare, ngakolunye uhlangothi, ikulungiselele konke. Ubhala nje ikhodi ukuze uyibize, futhi iphatha zonke izinto ezingcolile kuwe.

Ngicabanga ukuthi lo mqondo wokucabanga ulungile. Amakhono e-AI azokhula abe njengengqalasizinda efana namanzi nogesi; awudingi ukwazi ukuthi ugesi ukhiqizwa kanjani, ungawuxhuma nje bese uwusebenzisa. I-Cloudflare Workers AI ithatha lo mqondo "wokuxhuma nokudlala" iwuyise ezingeni elibi kakhulu.

Vele, ayinazo izinkinga zayo. Isilinganiso samahhala singama-Neuron ayi-10000 kuphela, ngakho-ke kuzodingeka ukhokhe uma ufuna ukusifaka endaweni yokukhiqiza enethrafikhi ephezulu. Futhi, yize kunezinhlobo ezingaphezu kuka-50, uma kuqhathaniswa ne-ecosystem ephelele ye-OpenAI, amanye amamodeli amancane noma amasha awasekelwa.

Kodwa ngamaphrojekthi omuntu siqu, amathuluzi amancane, noma nje izivivinyo eziqhutshwa ilukuluku, ngicabanga ukuthi kwanele.

Yilokho engikubonile ngempela. Njengoba usufundile kuze kube manje, uma ukuthole kuwusizo, sicela uthande futhi wabelane ngakho. Uma ufuna ukuthola izibuyekezo kuqala, ungangilandela futhi! ⭐

Ngiyabonga ngokufunda isihloko sami. Sizobonana ngesikhathi esizayo.