Ukubuyekezwa kwesiphakamiso sesibonelelo seSibonelelo Sokuhlobisa usebenzisa i-Amazon Bedrock

Izinhlangano ezingenzi nzuzo zihlola iziphakamiso zesibonelelo zibhekene nenselelo ebalulekile: Ukuhlanza ngamakhulu okufakwayo okunemininingwane, ngakunye okuhambisana nokufaneleka okuhlukile, ukukhomba imizamo ethembisayo kakhulu. Le nqubo eqinile, edla isikhathi ngokujwayelekile isinyathelo sokuqala kwinqubo yokuphatha onesibonelelo, okubalulekile ekudleni umthelela wezenhlalo onenjongo.

Ithimba le-AWS lezeMfihlo Yezenhlalo kanye nomthelela Iqembu lakha isixazululo esisha sokuqondisa isiphakamiso sesiphakamiso sesibonelelo kanye nokuhlola ukusebenzisa amakhono wemvelo wokucubungula izilimi (NLP) e-Amazon Bedrock. I-Amazon Bedrock iyinsizakalo ephethwe ngokuphelele ekuvumela ukuthi usebenzise ukukhetha kwakho kwamamodeli wesisekelo aphezulu (FMS) kusuka ezinkampanini ze-AI ezihamba phambili njenge-AI21 Labs, i-AI Engi, ukuqina, ukuqina kwe-API, kanye ne-Amazon nge-API eyodwa, kanye Ngesethi ebanzi yamakhono owadingayo ukuze wakhe izinhlelo zokusebenza ezikhiqizayo ze-AI ngokuphepha, ubumfihlo, kanye ne-AI enomthwalo wemfanelo.

Ngokomlando, izinhlelo zokusebenza ze-AWS Health Ukuqulwa Kwezempilo kwabuyekezwa ngesandla yiKomidi Lokubukeza. Kuthathe izinsuku eziyi-14 noma ngaphezulu umjikelezo ngamunye wazo zonke izinhlelo zokusebenza zibuyekezwe ngokugcwele. Ngokwesilinganiso, uhlelo luthole izinhlelo ezingama-90 umjikelezo ngamunye. Umjikelezo wohlelo lokusebenza lwe-ONS 2024 AWS I-Equity Equity Aquiticity Curcle etholwe izinhlelo eziyi-139, i-TORPORPORT ENGCWELEKILE KULULA KAKHULU. Bekuzothatha izinsuku ezingama-21 zeKomidi Lokubuyekezwa ukuze licubungule ngesandla lezi zinhlelo zokusebenza. Indlela ye-Amazon Bedrock Centrock egxilile yehlise isikhathi sokubuyekezwa kwezinsuku ezi-2 (ukuncishiswa kwama-90%).

Injongo bekuwukuthuthukisa ukusebenza kahle kanye nokuvumelana kwenqubo yokubuyekezwa, ukunika amandla amakhasimende ukwakha izixazululo ezinomthelela ngokushesha okukhulu. Ngokuhlanganisa amakhono athuthukile we-NLP e-Amazon Bedrock ene-Defectiful Demplove Engineering, iqembu ladala isixazululo esishukumisayo, esiqhutshwa ngedatha, nesixazululo esilinganayo esibonisa amandla okuguqula amamodeli amakhulu (LLMS) esizindeni somthelela omkhulu.

Kulokhu okuthunyelwe, sihlola imininingwane yokuqalisa yezobuchwepheshe kanye nokufundwa okubalulekile okuvela kwi-Team's Amazon Bedrock Powered Power Grant Ukubuyekezwa Kokubukeza Isibonelelo, ukuhlinzeka nge-blueprint yezinhlangano ezifuna ukuthuthukisa izinqubo zokuphatha izibonelelo zawo.

Ukwakha isiphakamiso esisebenzayo sokubukeza iziphakamiso zesibonelelo usebenzisa i-ai ekhiqizayo

Ubunjiniyela obusheshayo ubuciko bokuqamba ama-Products asebenzayo ukufundisa nokuqondisa amamodeli we-AI wokwakha, njenge-LLMS, ukukhiqiza imiphumela oyifunayo. Ngokuklama ngokucabangayo okucatshangelwe, abasebenza ngodokotela bangavula amandla aphelele ezinhlelo ze-AI akhiqizayo futhi bazisebenzise ezinhlobonhlobo zezimo zangempela zomhlaba.

Lapho kwakhiwa imodeli ye-Amazon Bedrock yethu ukuze ibukeze iziphakamiso zesibonelelo, sasebenzisa amasu amaningi wobunjiniyela wokuqinisekisa ukuqiniseka ukuthi izimpendulo zemodeli zenzelwe, zihlelwe futhi zisebenze. Lokhu kufaka phakathi imodeli umuntu othile, enikeza imiyalo yesinyathelo ngesinyathelo, futhi acacise ifomethi yokuphuma oyifunayo.

Okokuqala, sabela i-Model the Persona yesazi empilweni yomphakathi, egxile ekuthuthukiseni imiphumela yokunakekelwa kwempilo kwabantu abangagciniwe. Lo mongo usiza i-PRIME Imodeli yokuhlola isiphakamiso ngokombono wesazi sendaba (SME) esicabanga ngokuholelisisa ngezinselelo zomhlaba wonke kanye nomthelela osezingeni lomphakathi. Ngokuchaza kahle i-Persona, siyaqiniseka ukuthi izimpendulo zemodeli zihambisana nelensi yokuhlola oyifunayo.

Your task is to review a proposal document from the perspective of a given persona, and assess it based on dimensions defined in a rubric. Here are the steps to follow:

1. Review the provided proposal document: {PROPOSAL}

2. Adopt the perspective of the given persona: {PERSONA}

Ama-Multiple Personas angabelwa ngokumelene ne-rubric efanayo ukuze aphendule ngemibono ehlukahlukene. Isibonelo, lapho i-Persona Public 'Nezempi Yezifundo Isazi Sokubaluleka, imodeli yahlinzeka ngokuqonda okujulile kumthelela omkhulu wephrojekthi amandla kanye nobufakazi. Lapho kwabelwa imali “yobukhulu bomuntu” Ngokufanayo, lapho unjiniyela we-software “lapho wabelwa khona, imodeli ehambisane nendaba yobuchwepheshe mayelana nokusetshenziswa kwe-AWS okuhlongozwayo.

Ngokulandelayo, sabhidliza inqubo yokubuyekezwa ibe seqoqo lemiyalo ehlelekile lemodeli okufanele liyilandele. Lokhu kufaka ukubukeza lesi siphakamiso, ukusihlonza ngobukhulu obuthile (amandla okuvutha, amakhono, kungenzeka, kungenzeka, ukuqina), bese unikeza isifinyezo sonke nesikolo. Ukuchaza le mikhombandlela yezinyathelo ngezinyathelo kunikeza isiqondiso esicacile semodeli ngezinto ezidingekayo zemisebenzi futhi zisiza ukukhiqiza ukuhlolwa okuphelele nokungaguquguquki.

3. Assess the proposal based on each dimension in the provided rubric: {RUBRIC}

For each dimension, follow this structure:

 
 Provide a brief summary (2-3 sentences) of your assessment of how well the proposal meets the criteria for this dimension from the perspective of the given persona. 
  Provide a score from 0 to 100 for this dimension. Start with a default score of 0 and increase it based on the information in the proposal. 
  Provide 2-3 specific recommendations for how the author could improve the proposal in this dimension. 


4. After assessing each dimension, provide an  section with:
 - An overall assessment summary (3-4 sentences) of the proposal's strengths and weaknesses across all dimensions from the persona's perspective
 - Any additional feedback beyond the rubric dimensions
 - Identification of any potential risks or biases in the proposal or your assessment

5. Finally, calculate the  by applying the weightings specified in the rubric to your scores for each dimension.

Ekugcineni, sachaza ifomethi yokuphuma oyifunayo njenge-JSS, enezingxenye ezihlukile zokuhlola okuyisici, okufingqiwe jikelele, kanye ne-Selive Scond. Ukunqunywa le fomethi yempendulo ehlelekile kwenza isiqiniseko sokuthi imiphumela yemodeli ingafakwa, igcinwe, futhi ihlaziywe yiqembu lethu lokuBukeza lesibonelelo, kunokuba lilethwe embhalweni wamafomu mahhala. Leli zinga lokulawula ngokuphuma lisiza ukuqondisa ukusetshenziswa kwe-Downstream kokuhlolwa kwemodeli.

6. Return your assessment in JSON format with the following structure:

{{ "dimensions": [ {{ "name": "", "summary": "
", "score": , "recommendations": [ "", "", ... ] }}, ... ], "overall_summary": "","overall_score":  }}

Do not include any other commentary beyond following the specified structure. Focus solely on providing the assessment based on the given inputs.

Ngokuhlanganisa lezi zindlela zobunjiniyela ezisheshayo – Indima yesandla, isabelo-sinyathelo ngesinyathelo, nokufometha kokukhipha – sakwazile ukwenza izikebhe ngokushesha okutholakele okuphelele, inhloso, kanye nokuhlolwa kwesiphakamiso sesibonelelo esisebenzayo esivela kumodeli yethu ye-AI ekhiqizayo. Le ndlela ehlelekile isenza sikwazi ukusebenzisa kahle amakhono wemodeli ukusekela inqubo yethu yokubuyekezwa yesibonelelo ngendlela esetshenziswayo futhi ephumelelayo.

Ukwakha uhlelo lokubuyekezwa kwesiphakamiso esishukumisayo nge-Streamlit and Maderative AI

Ukukhombisa nokuhlola amakhono esixazululo sokubuyekezwa kwesiphakamiso esishukumisayo, sakha ukusebenza okusheshayo kwe-prototype sisebenzisa ukusakaza, umbhede we-amazon, kanye ne-amazon dynanom. Kubalulekile ukuqaphela ukuthi lokhu kuqaliswa akuhloselwe ukusetshenziswa kokukhiqizwa, kepha kunalokho kusebenza njengobufakazi bomqondo kanye nendawo yokuqala ukuze kuthuthukiswe okunye. Uhlelo lokusebenza luvumela abasebenzisi ukuthi bachaze futhi balondoloze ama-rubrikhi ahlukahlukene kanye nama-rubrikhi ahlukahlukene, angasetshenziswa ngokuqinile lapho ebukeza izethulo zokuthumela. Le ndlela inika amandla ukuhlolwa okuhambisana nokufanele kwesiphakamiso ngasinye, ngokuya ngemibandela ebekiwe.

Isakhiwo sohlelo lokusebenza siqukethe izinto ezimbalwa ezibalulekile, esixoxa ngazo kulesi sigaba.

Iqembu lasebenzisa i-dynanom, i-database ye-NOSQL, ukugcina ama-personas, ama-rubriki, futhi neziphakamiso ezilethwe. Imininingwane egcinwe ku-DynanomDB yathunyelwa ukuqondisa, isikhombimsebenzisi sohlelo lokusebenza lweWebhu. Ku-Streamlit, iqembu lengeze i-Persona kanye ne-rubric to the esheshayo futhi lithumele i-templet to amazon bedrock.

import boto3
import json

from api.personas import Persona
from api.rubrics import Rubric
from api.submissions import Submission

bedrock = boto3.client("bedrock-runtime", region_name="us-east-1")

def _construct_prompt(persona: Persona, rubric: Rubric, submission: Submission):
    rubric_dimensions = [
        f"{dimension['name']}|{dimension['description']}|{dimension['weight']}"
        for dimension in rubric.dimensions
    ]

    # add the table headers the prompt is expecting to the front of the dimensions list
    rubric_dimensions[:0] = ["dimension_name|dimension_description|dimension_weight"]
    rubric_string = "n".join(rubric_dimensions)
    print(rubric_string)

    with open("prompt/prompt_template.txt", "r") as prompt:
        prompt = prompt.read()
        print(prompt)
        return prompt.format(
            PROPOSAL=submission.content,
            PERSONA=persona.description,
            RUBRIC=rubric_string,
        )

I-Amazon Bedrock yasebenzisa i-Anthropic's Claude 3 Sonnet FM ukuthi ihlole iziphakamiso ezifakiwe ngokumelene ne-Prompt. Imodeli yemodeli yakhiqizwa ngamandla ngokususelwa kumuntu okhethiwe kanye nerubriki. I-Amazon Bedrock ibizothumela imiphumela yokuhlola ibuyele ukuqondiswa kokubuyekezwa kweqembu.

def get_assessment(submission: Submission, persona: Persona, rubric: Rubric):
    prompt = _construct_prompt(persona, rubric, submission)

    body = json.dumps(
        {
            "anthropic_version": "",
            "max_tokens": 2000,
            "temperature": 0.5,
            "top_p": 1,
            "messages": [{"role": "user", "content": prompt}],
        }
    )
    response = bedrock.invoke_model(
        body=body, modelId="anthropic.claude-3-haiku-20240307-v1:0"
    )
    response_body = json.loads(response.get("body").read())
    return response_body.get("content")[0].get("text")

Umdwebo olandelayo ukhombisa umbukiso wesibalo esandulele.

Ukuchitheka komsebenzi kuqukethe lezi zinyathelo ezilandelayo:

1. Abasebenzisi bangadala futhi baphathe ama-Personas kanye nama-rubriki ngohlelo lokusebenza lwe-Streamlit. Lokhu kugcinwe kwi-DynamoDB Database.
   
   
2. Lapho umsebenzisi ehambisa isiphakamiso sokubuyekezwa, bakhetha umuntu ofisa kanye ne-rubric kusuka ezinketho ezitholakalayo.
   
3. Uhlelo lokusebenza lwe-Streamlit lukhiqiza ngokushesha okushukumisayo kwemodeli ye-Amazon Bedrock, okufaka imininingwane ekhethiwe ne-rubric imininingwane.
4. Imodeli ye-Amazon Bedrock ihlole isiphakamiso esekwe ekushesheni okunamandla futhi ibuyisa imiphumela yokuhlola.
5. Imiphumela yokuhlola igcinwe kwi-DynamoDB database futhi yethulwe kumsebenzisi ngohlelo lokusebenza lwe-Streamlit.
   

Umphumela

Le prototype esheshayo ikhombisa amandla enqubo yokubuyekezwa kwesiphakamiso esilinganiselwe futhi eguquguqukayo, evumela izinhlangano ukuba:

• Yehlisa isikhathi sokucubungula kwesicelo nge-90%
• Sakanisa inqubo yokubuyekezwa ngokushintsha imisebenzi yokuhlola
• Bamba idatha ehlelekile eziphakanyisiwe kanye nokuhlolwa kokuhlaziywa okwengeziwe
• Faka imibono ehlukahlukene ngokunika amandla ukusetshenziswa kwama-Multic amaningi namarubrucric

Kulo lonke ukuqaliswa, iqembu le-AWS SRI ligxile ekwakheni isipiliyoni esisebenzayo nesisebenziseka kalula. Ngokusebenza ngezandla-sisebenza ngohlelo lokusebenza lwe-Streamlit futhi ubheka umthelela we-Dynamic Persona kanye nokukhethwa kwe-rubric, abasebenzisi bangathola okuhlangenwe nakho okusebenzayo ekwakheni izinhlelo zokusebenza ezinamandla ze-AI ezikhuluma ngezinselelo zangempela zomhlaba.

Ukucatshangelwa kokusetshenziswa kwebanga lokukhiqiza

Yize i-prototype esheshayo ikhombisa amandla alesi sixazululo, ukusetshenziswa kwebanga lokukhiqiza kudinga ukucatshangelwa okwengeziwe kanye nokusetshenziswa kwezinsizakalo ezengeziwe ze-AWS. Okunye ukucatshangelwa okubalulekile kufaka phakathi:

• Ukukala nokusebenza – Ngokuphatha amavolumu amakhulu eziphakamiso kanye nabasebenzisi bakwa-Conferent, Isakhiwo esingenama-AWS Lambda, i-Amazon API Gateway, ne-Amazon Storage Service (I-Amazon S3) ibizohlinzeka ngezikali okungcono, ukutholakala, kanye nokwethembeka.
• Ezokuphepha Nokuhambisana – Kuya ngokuzwela kwemininingwane ehilelekile, izindlela ezengeziwe zokuphepha ezinjengokubethela, ubuqiniso kanye nokulawulwa kokufinyelela, kanye nokuhlolwa kwamabhuku kuyadingeka. Izinsizakalo ezinjenge-AWS Key Management Service (KMS), i-Amazon Cognito, ubunikazi be-AWS kanye nokuphathwa kokufinyelela (IAM), kanye nama-AWS Clubrail angasiza ukufeza lezi zidingo.
• Ukuqapha nokungena ngemvume – Ukuqalisa ukuqapha okuqinile nokuqashwa kwama-loging kusetshenziswa izinsizakalo ezifana ne-Amazon CloudWatch kanye ne-AWX X-Ray Nika amandla okulandela umkhondo wokulandela umkhondo, ukukhomba izingqinamba, nokugcina ukuhambisana.
• Ukuhlolwa okuzenzakalelayo nokuthunyelwa – Ukuqalisa Ukuhlolwa kokuzenzakalelayo nokuthumela amapayipi kusetshenziswa izinsizakalo ezinjenge-AWS CodePipeline, Usizo lwe-AWS Codedeplond lula ukuhlinzekwa okungaguquki nokungaguquki, ukunciphisa ubungozi bamaphutha nesikhathi sokuphumula.
• Ukwenziwa kwezindleko – Ukuqalisa ukusebenzisa amasu wokusebenziseka kwezindleko, njengokusebenzisa ama-AWS abiza ama-Explorer nama-AWS wezabelomali, kungasiza ukuphatha izindleko futhi kusiza ukugcina ukusetshenziswa kwezinsiza kusebenza kahle.
• Ukucatshangelwa kwe-AI enomthwalo wemfanelo – Ukuqalisa Ukuvikela – Njenge-Amazon Bedrock Guardrails-kanye Nezindlela Zokuhlola kungasiza ekuphoqeleleni ukusetshenziswa okuphathelene nokuziphatha kwemodeli ye-AI ekhiqizayo, kufaka phakathi ukutholwa kwe-bias, kufaka phakathi ukutholwa kokudla, nokuhlolisisa komuntu, kanye nokwengamela komuntu, kanye nokwengamela komuntu. Yize ifomu lesicelo sempilo se-AWS liqoqiwe imininingwane yamakhasimende afana negama, ikheli le-imeyili, kanye nezwe lokusebenza, lokhu kushiywe ngokuhlelekile lapho kuthunyelwa kuthuluzi le-Amazon Bedrock elinike amandla ukugwema ukukhetha kumodeli futhi kuvikele idatha yamakhasimende.

Ngokusebenzisa i-suite ephelele yezinsizakalo ze-AWS futhi kulandela izindlela ezingcono kakhulu zokuphepha, ukuqina, nezinhlangano ze-AI, zingakha isixazululo esilungele ukukhiqizwa esihlangabezana nezidingo ezithile ngenkathi sithola ukuhambisana, ukuthembeka kanye nokusebenza kwezindleko.

Ukugcina

I-Amazon Bedrock-ehambisana nama-AWS asebenzayo ama-AWS anikeze amandla okubukeza iziphakamiso zesibonelelo futhi alethe imiklomelo kumakhasimende ngezinsuku esikhundleni samasonto. Amakhono athuthukiswe kulo msebenzi – njengezicelo zeWebhu ezinjengezilungiselelo ze-Sysklit, ezihlanganisa nolwazi lwe-NoSQL njengeDynamoDB, futhi ngokwezifiso ezikhiqizwayo ze-AI ze-AI futhi zisebenza ezinhlobonhlobo zezimboni futhi zisebenzisa amacala.

Mayelana nababhali

UCarolyn Vigil kungumsebenzi womhlaba wonke wemibandela yokuzibophezela kwezenhlalo kanye nomthelela wokubambisana kwamakhasimende. Ushayela imizamo yamasu atholwe amafu we-Computer for umthelela emphakathini emhlabeni wonke. Ummeli onentshisekelo yemiphakathi engagciniwe, uphinde wasungula izinhlangano ezimbili ezingenzi nzuzo ezisebenzela abantu abakhubazekile ngentuthuko nemindeni yabo. UCarolyn uthokozela ukufika kwentaba nomndeni wakhe nabangane ngesikhathi sakhe samahhala.

Lauren Hollis Ingabe Umphathi Wezinhlelo Zokuzibophezela Komphakathi Womphakathi Womphakathi kanye nomthelela. Ubeka isizinda sakhe kwezomnotho, ucwaningo lwezempilo, kanye nobuchwepheshe ukusekela izinhlangano eziqhutshwa yimisebenzi ziletha umthelela wenhlalo usebenzisa ubuchwepheshe befu lama-AWS. Ngesikhathi sakhe samahhala, uLauren uyakujabulela ukufunda upiyano no-cello.

Ben West Ingabe umakhi wezandla onesipiliyoni ekufundeni komshini, ama-analytics amakhulu wedatha, kanye nokuthuthukiswa kwesoftware okugcwele. NjengoMphathi Wezobuchwepheshe Wezobuchwepheshe ku-AWS Social Betformation Team & Umthelela, uBen ubekezelela izinhlobo eziningi zamafu, onqenqemeni, ne-inthanethi yezinto (iot) ubuchwepheshe ukuthuthukisa ama-prototypes amasha futhi asize izinhlangano zomkhakha womphakathi zenza umthelela omuhle emhlabeni. UBen ungumthengisi wezempi ojabulela ukupheka nokuba ngaphandle.

Mike Haggerty Ingabe unjiniyela wokuthuthuka kwamaSysite (Sr. Sysde) e-Amazon Web Services (AWS), esebenza ngaphakathi kweqembu le-pace-odge. Kule ndima, uneqhaza emizani ye-aw uso se-compling njengengxenye yeqembu lomkhakha womphakathi womphakathi (we-WWPS) weqembu lomhlaba wonke (ama-prototyping kanye nezobunjiniyela bezezimali). Ngale kwemisebenzi yakhe yobungcweti, uMike ungumhlinzeki wezilwane ozithandweni, kanye nenja yakhe i-Gnocchi, ihlinzeka ngezinsizakalo zokusekelwa ezikhungweni zomphakathi wendawo.