Yakha uhlelo lokuphatha izindleko lwe-AI Fors Management for Amazon Bedrock – Ingxenye 2

Engxenyeni 1 yochungechunge lwethu, sethula isixazululo sokuphathwa kwezindleko okusebenzayo kwe-Amazon Bedrock, enesikhwama sezinkambiso esinamandla esenzelwe ukusebenzisa umkhawulo wokusetshenziswa kwe-TEAL-time. Sihlole amasu wezakhiwo ezibalulekile, amasu wokulandela amathokheni, kanye namasu okusebenza kwesabelomali sokuqala esisiza izinhlangano ukulawula izindleko zabo ze-AI ezikhiqizayo.

Ukwakha kuleso sisekelo, lokhu okuthunyelwe kuhlola amasu wokuqapha izindleko ezithuthukile zokuhanjiswa kwe-AI sebekhiqizayo. Sethula izindlela zokumaka ngokwesiko le-granular ukuze uthole ukwabiwa kwezindleko okuqondile, futhi sithuthukise izindlela zokubika eziphelele.

Ukubuka konke

Isixazululo se-Sedry Forts yethulwe engxenyeni 1 yathuthukiswa njengendlela ephakathi nendawo yokukhawulela ukukhiqizwa kwe-AI generative ake ukunamathela kwizabelomali ezibekiwe. Umdwebo olandelayo ukhombisa izingxenye eziyinhloko zesisombululo, wengeza ngokuqapha izindleko ngokusebenzisa ukukhokhiswa kwama-AWS kanye nokuphathwa kwezindleko.

Ukumaka kwezinga le-invocation

Ukumaka kwezinga le-invocation kunwebisa amakhono ethu wekhambi ngokunamathisela iMetadata ecebile kuzo zonke izicelo ze-API, kwakha umzila obanzi wokuhlola ngaphakathi kwamalogi we-Amazon Cloudwatch. Lokhu kuba yigugu ikakhulukazi lapho kuphenya izinqumo ezihlobene nesabelomali, kuhlaziya imithelela emincane kakhulu, noma ukuqonda amaphethini wokusebenzisa kuzo zonke izinhlelo namaqembu ahlukene. Ukusekela lokhu, imisebenzi ephambili yezinyathelo ze-AWS Workflow yavuselelwa, njengoba kuboniswe kulokhu okulandelayo.

Ukufakwa kwe-API okuthuthukisiwe

Siphinde sathola okokufaka kwe-API ukusekela ukumaka ngokwezifiso. Isakhiwo esisha sokufaka sethula amapharamitha ongazemukela ukucushwa okuthile okucacisiwe kanye nokumaka ngokwezifiso:

{
  "model": "string",     // e.g., "claude-3" or "anthropic.claude-3-sonnet-20240229-v1:0"
  "prompt": {
    "messages": [
      {
        "role": "string",    // "system", "user", or "assistant"
        "content": "string"
      }
    ],
    "parameters": {
      "max_tokens": number,    // Optional, model-specific defaults
      "temperature": number,   // Optional, model-specific defaults
      "top_p": number,         // Optional, model-specific defaults
      "top_k": number          // Optional, model-specific defaults
    }
  },
  "tags": {
    "applicationId": "string",  // Required
    "costCenter": "string",     // Optional
    "environment": "string"     // Optional - dev/staging/prod
  }
}

Isakhiwo sokufaka sinamazakhi ezintathu ezibalulekile:

• isifanekiso – Amamephu alula amagama (ngokwesibonelo, claude-3) Kumazisi ezigcwele ze-Amazon Bedrock Model (ngokwesibonelo, anthropic.claude-3-sonnet-20240229-v1:0Isihlehlukene
• faka – Inikeza ama-ARAMS ARAMS for Promples, Ukusekela Kokubili I-Single-Vula Izingxoxo Nezingxoxo Eziningi
• Amathegi – Usekela ukulandelwa kwezinga lesicelo, nge applicationId njengenkambu edingekayo futhi costCenter na- environment Njengamasimu wokuzikhethela

Kulesi sibonelo, sisebenzisa izikhungo zezindleko ezahlukahlukene ze sales, servicesfuthi support Ukuze kulingise ukusetshenziswa kwemfanelo yebhizinisi ukulandelela ukusetshenziswa kanye nokusebenzisa ukutholwa kwe-Amazon Bedrock. Ngokwesibonelo:

{
  "model": "claude-3-5-haiku",
  "prompt": {
    "messages": [
      {
        "role": "user",
        "content": "Explain the benefits of using S3 using only 100 words."
      },
      {
        "role": "assistant",
        "content": "You are a helpful AWS expert."
      }
    ],
    "parameters": {
      "max_tokens": 2000,
      "temperature": 0.7,
      "top_p": 0.9,
      "top_k": 50
    }
  },
  "tags": {
    "applicationId": "aws-documentation-helper",
    "costCenter": "support",
    "environment": "production"
  }
}

Ukuqinisekiswa nokumaka

Kwangezwa isinyathelo esisha sokuqinisekiswa ekuhambeni komsebenzi ngokumaka. Lesi sinyathelo sisebenzisa umsebenzi we-AWS Lambda ukwengeza amasheke wokuqinisekiswa namamephu Imodeli ecelwe i-ID ethile yemodeli e-Amazon Bedrock. Kufinyelela tags into enamathegi azodingeka ekuhlaziyweni okuphansi.

Le khodi elandelayo iyisibonelo semephu elula yokuthola i-ID yemodeli efanelekile kusuka kumodeli echaziwe:

MODEL_ID_MAPPING = {
    "nova-lite": "amazon.nova-lite-v1:0",
    "nova-micro": "amazon.nova-micro-v1:0",
    "claude-2": "anthropic.claude-v2:0",
    "claude-3-haiku": "anthropic.claude-3-haiku-20240307-v1:0",
    "claude-3-5-sonnet-v2": "us.anthropic.claude-3-5-sonnet-20241022-v2:0",
    "claude-3-5-haiku": "us.anthropic.claude-3-5-haiku-20241022-v1:0"
}

Ukungena ngemvume nokuhlaziywa

Ngokusebenzisa amame metrics ama-metric anama-tag akhiqizwe ama-Cloud Amathegi ngokwezifiso nobukhulu bakhomba ukuthi amaqembu asebenzisa kanjani izinsizakalo ze-AI. Ukubona lokhu kuhlaziya, kwaqaliswa izinyathelo zokukhiqiza amathegi e-Metric, i-Store metric, futhi ihlaziye idatha ye-metric:

1. Sifaka isethi eyingqayizivele yamathegi athwebula imininingwane yokuqukethwe. Lokhu kungafaka amathegi ahlinzekwe ngabasebenzisi kanye nalezo ezikhiqizwa ngamandla, njenge requestId na- timestamp:

     "tags": {
       "requestId": "ded98994-eb76-48d9-9dbc-f269541b5e49",
       "timestamp": "2025-01-31T14:05:26.854682",
       "applicationId": "aws-documentation-helper",
       "costCenter": "support",
       "environment": "production"
   }

2. Njengoba ukugeleza komsebenzi ngamunye kubulawa, kuzohlolwa umkhawulo wemodeli ngayinye ukuze uqiniseke ukuthi isicelo singaphakathi kwemihlahlandlela yesabelomali. Ukuhamba komsebenzi kuzophela ngokuya ngemiphumela emithathu engenzeka:
   1. I-Rate Limit ivunyelwe futhi i-invationsuration iphumelele
   2. Ukulinganisa umkhawulo kuvunyelwe futhi ukunxusa kungaphumelelanga
   3. Umkhawulo wokulinganisa uyakwenqatshwa

   Idatha ye-metric yangokwezifiso igcinwa ku-CloudWatch ku GenAIRateLimiting igama elithi. Le namespace ifaka amamethrikhi asemqoka alandelayo:

   • I-TOALLREQESS – Kubalwa wonke umzamo wokuncenga kungakhathalekile imiphumela
   • Hle – Amathrekhi izicelo ezidlulise amasheke okukhawulelwa
   • Ratelimitded – Amathrekhi izicelo ezivinjelwe ngokukhawulelwa kwamanani
   • UkuncishiswaFaked – Kubalwa izicelo ezihlulekile ngesikhathi sokucela imodeli
   • Kufakwa okufakiwe – Ukusetshenziswa kokufaka okufakwayo kwezicelo eziphumelelayo
   • Okuphumayo – Izinyathelo zokuthenga zethokheni eziphuma ezicelweni eziphumelelayo

   I-metric ngayinye ifaka ubukhulu be Model, ModelId, CostCenter, Applicationfuthi Environment kokuhlaziywa kwedatha.

3. Sisebenzisa ama-coffiwatch metryc metric mathrics amakhono ngezindlela zezibalo zokuhlaziya idatha eqoqwe yi-Workflow. Imininingwane ingakhonjiswa kumafomethi ahlukahlukene abukwayo ukuthola umbono we-granular wezicelo ngobukhulu obunikezwe, njengemodeli noma isikhungo sezindleko. Isikrini esilandelayo sibonisa ideshibhodi eyisibonelo ekhombisa amamethrikhi we-inforation lapho imodeli eyodwa isifinyelele umkhawulo wayo.

Ama-Amazon Bedrock Analytics

Ngokungeziwe kumadeshibhodi e-metric akwenziwe ngokwezifiso, amafushiwatch ahlinzeka ngamadeshi azenzakalelayo wokuqapha ukusebenza kwe-Amazon Bedrock nokusetshenziswa. Le khasi Umbhede I-Dashboard inikeza ukubonakala kumamethrikhi asemqoka kanye nokuqonda okusebenzayo, njengoba kukhonjisiwe ku-skrini elandelayo.

Ukumaka izindleko nokubika

I-Amazon Bedrock yethule amaphrofayili wokuhlaselwa kohlelo lokusebenza, amandla amasha izinhlangano angawasebenzisa ukufaka amathegi ekwabiwa kwezindleko ukulandelela nokuphatha ukusetshenziswa kwemodeli yazo (FM) yazo. Lesi sici sibheka umkhawulo wangaphambilini lapho ukumaka bekungenzeki khona ku-FMS efunwayo, okwenza kube nzima ukulandelela izindleko kuwo wonke amayunithi ebhizinisi ahlukene kanye nezinhlelo zokusebenza. Manje usungadala amaphrofayili alandelako ama-FMS we-FMS bese usebenzisa amathegi isabelo sokwabiwa kwezindleko njengomnyango, iqembu kanye nezikhombi zohlelo lokusebenza. Lawa ma-tag ahlanganisa namathuluzi wokuphathwa kwezindleko ze-AWS kufaka phakathi ama-AWS abiza umhloli wamazwe, amasabelomali amasha, kanye nama-AWS abiza ukutholwa kwe-anomaly, ukunika amandla ukuhlaziywa kwezindleko okuningiliziwe kanye nokulawulwa kwesabelomali.

Amaphrofayili wokufaka isicelo

Ukuze uqale, kufanele udale amaphrofayili okulandela uhlelo lokusebenza ngohlobo ngalunye lokusetshenziswa ofuna ukulandelela. Kulokhu, ikhambi lichaza amathegi wangokwezifiso costCenter, environmentfuthi applicationId. Iphrofayili yokuvumisana nayo izokwesuselwa kuphrofayela yemodeli ye-Amazon Bedrock ekhona, ngakho-ke kufanele uhlanganise amathegi afunekayo nemodeli kuphrofayela. Ngesikhathi sokubhala, kufanele usebenzise i-AWS Command Line Interface (AWS CLI) noma ama-AWS API ukudala eyodwa. Bona ikhodi eyisibonelo elandelayo:

aws bedrock create-inference-profile 
  --inference-profile-name "aws-docs-sales-prod" 
  --model-source '{"copyFrom":  "arn:aws:bedrock:us-east-1::foundation-model/anthropic.claude-3-haiku-20240307-v1:0"}' 
  --tags '[
    {"key": "applicationId", "value": "aws-documentation-helper"},
    {"key": "costCenter", "value": "sales"},
    {"key": "environment", "value": "production"}
  ]'

Lo myalo udala iphrofayili yesikhungo sezindleko zokuthengisa kanye nemvelo yokukhiqiza usebenzisa imodeli ye-Anthropic's Claude Haiku 3.5. Ukukhishwa okuvela kulo myalo yigama lesisetshenziswa se-Amazon (ARN) ozolisebenzisa njenge-ID yemodeli. Kulesi sixazululo, ValidateAndSetContext Umsebenzi weLambda ushintshwe ukuze uvumele ukucacisa imodeli ngesikhungo sezindleko (ngokwesibonelo, sales). Ukubona ukuthi yimaphi amaphrofayili owadalile, sebenzisa umyalo olandelayo:

aws bedrock list-inference-profiles --type-equals APPLICATION

Ngemuva kokuthi amaphrofayli adaliwe futhi ukuqinisekiswa kubuye kwavuselelwa ezikhungweni zezindleko zemephu kuma-arns wephrofayili, ukugeleza komsebenzi kuzoqala ukusebenzisa izicelo zokuhlobisa ngephrofayili ehambisanayo. Isibonelo, lapho umsebenzisi ehambisa isicelo, azocacisa imodeli njenge sales, servicesnoma support ukuvumelanisa nezikhungo ezintathu zezindleko ezichazwe. Le khodi elandelayo iyimephu efanayo esibonelweni esedlule:

MODEL_ID_MAPPING = {
    "sales": "arn:aws:bedrock:::application-inference-profile/",
    "services": "arn:aws:bedrock:::application-inference-profile/",
    "support": "arn:aws:bedrock:::application-inference-profile/"
   }

Ukubuza amamethrikhi we-CloudWatch ngokusetshenziswa kwemodeli kahle lapho usebenzisa amaphrofayili akwa-infence wokufaka isicelo, kufanele usho i-ID eyingqayizivele yephrofayili (ingxenye yokugcina ye-ARN). I-CloudWatch izogcina amamethrikhi afana nokusetshenziswa kwethokheni ngokuya nge-ID eyingqayizivele. Ukusekela womabili amaphrofayili kanye nokusetshenziswa kwemodeli eqondile, umsebenzi we-lambda waguqulwa ukwengeza ithegi entsha ye modelMetric ukuba yigama elifanele ukulisebenzisa ukubuza ukusetshenziswa kwethokheni. Bona ikhodi elandelayo:

  "tags": {
    "requestId": "ded98994-eb76-48d9-9dbc-f269541b5e49",
    "timestamp": "2025-01-31T14:05:26.854682",
    "applicationId": "aws-documentation-helper",
    "costCenter": "support",
    "environment": "production",    
    "modelMetric": " | "
  }

Ibiza Explorer

I-Cost Explorer iyithuluzi elinamandla lokuphathwa kwezindleko elihlinzeka ngamehlo aphelele nokuhlaziywa kwemali yakho yokusebenzisa imali kuma-AWS Services, kufaka phakathi i-Amazon Bedrock. Inikeza amadeshibhodi enembile ukulandelela izindleko zomlando, ukubikezela izindleko zesikhathi esizayo, futhi uthole imininingwane ekusetshenzisweni kwakho kwefu. Nge-FAST Explorer, ungaphula izindleko ngensizakalo, omaki, nobukhulu obujwayelekile, ukuze uthole ukuhlaziywa okuningiliziwe kwezezimali. Ithuluzi livuselelwa nsuku zonke.

Uma usebenzisa amaphrofayili wokuhlanza wohlelo lokusebenza nge-Amazon Bedrock, ukusetshenziswa kwensiza yakho ye-AI kuyamakwa ngokuzenzakalelayo futhi kugeleza ngokuzenzakalelayo ekukhokhweni nokuphathwa kwezindleko. Lawa ma-tags anika amandla okulandelwa kwezindleko okuningiliziwe ngezilinganiso ezihlukile njenge-Cost Center, uhlelo lokusebenza, kanye nemvelo. Lokhu kusho ukuthi ungakwazi ukukhiqiza imibiko ehlehla izindleko ze-Amazon Bedrock AI ngamayunithi athile webhizinisi, amaphrojekthi, noma ama-hierarchies ahlelekile, enikeza ukubonakala okucacile ekusetshenzisweni kwemali yakho ekhiqizayo ye-AI.

Amathegi ekwabiwa kwezindleko

Amathegi ekwabiwa kwezindleko ngamabili amanani aphezulu asiza ngezigaba futhi ulandelele izindleko zezindleko ze-AWS enhlanganweni yakho. Ngokwesimo se-Amazon Bedrock, la ma-tag angafaka izimfanelo njengegama lesicelo, isikhungo sezindleko, imvelo, noma i-ID yephrojekthi. Ukuze usebenzise ithegi yokwabiwa kwezindleko, kufanele uqale uwunike amandla kwikhonsoli yokukhokha kanye nokuphathwa kwezindleko. Ngemuva kokuthi icushiwe, lawa ma-tag azovela kwizindleko zakho ze-AWS kanye noMbiko Wokusetshenziswa (RECR), Ukusiza Uphule izindleko ze-Amazon Bedrock ngemininingwane ye-granular.

Ukuze usebenzise ithegi yokwabiwa kwezindleko, qedela lezi zinyathelo ezilandelayo:

1. On the ukukhokhiswa kanye nekhonsoli ukuphathwa kwezindleko, efeni lokuhambisa, khetha Amathegi ekwabiwa kwezindleko.
2. Thola umaki wakho (ngalesi sibonelo, igama costCenter) bese ukhetha Vuma ukwenza.
3. Qinisekisa ukusebenza.

Ngemuva kokusebenza, costCenter I-Tag izovela ku-RUS yakho futhi izosetshenziswa ngokuhloliseka kwezindleko. Kungathatha amahora angama-24 ukuthi ithegi isebenze ngokuphelele emibikweni yakho yokukhokha.

Ukubika kwe-Fort Explorer

Ukwakha umbiko wokusetshenziswa kwe-Amazon Bedrock nge-Cost Explorer ngokususelwa kuthegi yakho, Qedela lezi zinyathelo ezilandelayo:

1. Kwikhonsoli yokuphatha yokukhokha kanye nokubiza, khetha Ibiza Explorer kufasitelana lokuzulazula.
2. Setha uhla lwakho olufisayo (ububanzi besikhathi esihlobene noma isikhathi esingokwezifiso).
3. Qoka -Zinsuku zonke noma -Nyanga zonke granularity.
4. Use Iqembu ngo Imenyu eyehlayo, khetha Umgoga.
5. Qoka costCenter njengokhiye wethegi.
6. Buyekeza izindleko eziboniswe i-Amazon Bedrock izindleko eziphulwe phansi ngenani ngalinye lesikhungo sezindleko.
7. Ngokuzithandela, hlunga amanani ngokusebenzisa isihlungi ku Isihlungi ingxenye:
   1. Qoka Umgoga Hlunga.
   2. Khetha I-Castcenter tag.
   3. Khetha amanani wesikhungo sezindleko ezithile ofuna ukusihlaziya.

Umbiko ophumayo uzohlinzeka ngombono onemininingwane nge-Amazon Bedrock AI Service izindleko, ukukusiza ukuqhathanisa ukusetshenziswa kwemali kumayunithi noma amaphrojekthi ahlukahlukene wenhlangano ngokunemba.

Ukubeka kafushane

Izindleko ze-AWS kanye nemibiko yokusebenzisa (kufaka phakathi isabelozimali) isenzo njengezinkomba zomgwaqo onqenqemeni ngoba zikhombisa ukuthi yini osuvele uyichithe e-Amazon Bedrock ngemuva kweqiniso. Ngokuxuba izexwayiso zesikhathi sangempela kusuka emisebenzini yegreyithi enemibiko ephelele yezindleko, ungathola ukubuka okungu-360-degree kokusetshenziswa kwe-Amazon Bedrock. Lokhu kubikwa kungakwazisa ngaphambi kokubheka ngokweqile futhi kukusize uqonde ukusetshenziswa kwakho kwangempela. Le ndlela ikunikeza amandla okuphatha izinsiza ze-AI ezisebenzayo, ukugcina isabelomali sakho esisha esilandelweni futhi amaphrojekthi akho asebenza kahle.

Zama le ndlela yokuphathwa kwezindleko yecala lakho lokusebenzisa, bese wabelana ngempendulo yakho kumazwana.

Mayelana nomlobi

Jason salcido Ingabe ukwakhiwa kwezixazululo eziphakeme zokuqalisa neminyaka ecishe ibe ngu-30 yokuhlangenwe nakho okuphayona izixazululo ezintsha zezinhlangano kusuka ekuqaleni kuya kumabhizinisi. Ubuchwepheshe bakhe buchitha ubuciko befu, i-computer engenasici, ukufunda ngomshini, ukukhiqizwa kwe-AI, nezinhlelo ezisatshalaliswa. UJason uhlanganisa ulwazi lobuchwepheshe olujulile ngendlela yokucabanga phambili yokucabanga ukuklama izixazululo ezi-scaleble ezishayela inani, ngenkathi uhumusha imiqondo eyinkimbinkimbi kumasu asebenzayo.