I-Generative Data Intelligence

AI

Izisekelo Zolwazi ze-Amazon Bedrock manje zisekela imiyalo yangokwezifiso ye-RetrieveAndGenerate API kanye nokucushwa kwenani eliphezulu lemiphumela ebuyisiwe | Izinsizakalo Zewebhu ze-Amazon

Ishicilelwe kabusha nguPlato
Ishicilelwe kabusha nguPlato

Usuku:

Ukubukwa: 0

With Izisekelo Zolwazi ze-Amazon Bedrock, ungaxhuma ngokuphephile amamodeli esisekelo (ama-FM) kuwo I-Amazon Bedrock kudatha yenkampani yakho ye-Retrieval Augmented Generation (RAG). Ukufinyelela kudatha eyengeziwe kusiza imodeli ukuthi ikhiqize izimpendulo ezifanele kakhulu, eziqonde ngqo, nezinembile ngaphandle kokuqeqesha kabusha ama-FM.

Kulokhu okuthunyelwe, sixoxa ngezici ezimbili ezintsha Zezisekelo Zolwazi ze-Amazon Bedrock eziqondene ne RetrieveAndGenerate I-API: ukulungisa inani eliphezulu lemiphumela nokudala ukwaziswa kwangokwezifiso ngesifanekiso sokwaziswa kwesisekelo solwazi. Manje usungakhetha lokhu njengezinketho zombuzo eduze nohlobo lokusesha.

Ukubuka konke kanye nezinzuzo zezici ezintsha

Inombolo ephezulu yenketho yemiphumela ikunikeza ukulawula inani lemiphumela yosesho ezobuyiswa esitolo se-vector futhi idluliselwe ku-FM ukuze ukhiqize impendulo. Lokhu kukuvumela ukuthi wenze ngendlela oyifisayo inani lolwazi lwangemuva olunikeziwe ukuze lukhiqize, ngaleyo ndlela unikeze umongo owengeziwe wemibuzo eyinkimbinkimbi noma ngaphansi emibuzweni elula. Ikuvumela ukuthi ulande imiphumela efika kweyi-100. Le nketho isiza ukuthuthukisa amathuba okuba komongo ofanele, ngaleyo ndlela ithuthukise ukunemba futhi inciphise ukubonwa kwempendulo ekhiqizwayo.

Isifanekiso sokwaziswa kwesisekelo solwazi ngokwezifiso sikuvumela ukuthi umiselele isifanekiso sokwaziswa esimisiwe ufake esakho ukuze wenze ngendlela oyifisayo ukwaziswa okuthunyelwa kumodeli ukuze kwenziwe impendulo. Lokhu kukuvumela ukuthi wenze ngendlela oyifisayo ithoni, ifomethi yokuphumayo, nokuziphatha kwe-FM uma iphendula umbuzo womsebenzisi. Ngale nketho, ungakwazi ukulungisa kahle amagama ukuze afane kangcono nomkhakha wakho noma isizinda (njengokunakekelwa kwezempilo noma okusemthethweni). Ukwengeza, ungakwazi ukwengeza imiyalelo yangokwezifiso nezibonelo ezihambisana nokugeleza komsebenzi wakho othize.

Ezigabeni ezilandelayo, sichaza ukuthi ungasebenzisa kanjani lezi zici nge- I-AWS Management Console noma i-SDK.

Okudingekayo

Ukuze ulandele lezi zibonelo, udinga ukuba nesisekelo solwazi esikhona. Ukuze uthole imiyalelo yokudala eyodwa, bheka Dala isisekelo solwazi.

Lungiselela inani eliphezulu lemiphumela usebenzisa ikhonsoli

Ukuze usebenzise inombolo enkulu yenketho yemiphumela usebenzisa ikhonsoli, qedela lezi zinyathelo ezilandelayo:

  1. Ku-console ye-Amazon Bedrock, khetha Izisekelo zolwazi ngakwesokunxele sokungena kwesokunxele.
  2. Khetha isisekelo solwazi osidalile.
  3. Khetha Isisekelo solwazi lokuhlola.
  4. Khetha isithonjana sokumisa.
  5. Khetha Vumelanisa umthombo wedatha ngaphambi kokuthi uqale ukuhlola isisekelo sakho solwazi.
  6. Ngaphansi Ukulungiselelwa, ngenxa Uhlobo losesho, khetha uhlobo lokusesha ngokusekelwe esimweni sakho sokusebenzisa.

Kulokhu okuthunyelwe, sisebenzisa usesho oluxubile ngoba luhlanganisa ukusesha kwe-semantic nombhalo ukuze kube nokunemba okukhulu komhlinzeki. Ukuze ufunde kabanzi mayelana nosesho lwe-hybrid, bona Izisekelo Zolwazi ze-Amazon Bedrock manje zisekela ukusesha okuyingxube.

  1. Nweba Inombolo enkulu yezingcezu zomthombo bese usetha inombolo yakho ephezulu yemiphumela.

Ukubonisa inani lesici esisha, sibonisa izibonelo zokuthi ungakhuphula kanjani ukunemba kwempendulo ekhiqiziwe. Sasebenzisa Idokhumenti ye-Amazon 10K yango-2023 njengomthombo wedatha wokudala isisekelo solwazi. Sisebenzisa lo mbuzo olandelayo ukuze sihlole: “Ngamuphi unyaka imali engenayo yonyaka ye-Amazon inyuke isuka ku-$245B yaya ku-$434B?”

Impendulo efanele yalo mbuzo ithi “Imali engenayo yonyaka ye-Amazon inyuke isuka ku-$245B ngo-2019 yaya ku-$434B ngo-2022,” ngokusekelwe emibhalweni esesisekelweni solwazi. Sisebenzise i-Claude v2 njenge-FM ukuze sikhiqize impendulo yokugcina esekelwe olwazini lwengqikithi olutholwe kusisekelo solwazi. UClaude 3 Sonnet kanye noClaude 3 Haiku nabo bayasekelwa njengama-FM esizukulwane.

Sisebenzise omunye umbuzo ukuze sibonise ukuqhathaniswa kokubuyisa ngokulungiselelwa okuhlukile. Sisebenzise umbuzo ofanayo wokufaka (“Ngamuphi unyaka imali engenayo yonyaka ye-Amazon inyuke isuka ku-$245B yaya ku-$434B?”) futhi setha inani eliphezulu lemiphumela laba ngu-5.

Njengoba kukhonjisiwe kusithombe-skrini esilandelayo, impendulo ekhiqiziwe yaba "Uxolo, angikwazi ukukusiza ngalesi sicelo."

Okulandelayo, sibeka imiphumela ephezulu ibe ngu-12 futhi sibuze umbuzo ofanayo. Impendulo ekhiqiziwe ithi "Imali engenayo yonyaka ye-Amazon isuka ku-$245B ngo-2019 yaya ku-$434B ngo-2022."

Njengoba kuboniswe kulesi sibonelo, siyakwazi ukubuyisa impendulo efanele ngokusekelwe enanini lemiphumela ebuyisiwe. Uma ufuna ukufunda okwengeziwe mayelana nesichasiso somthombo esakha okukhiphayo kokugcina, khetha Bonisa imininingwane yomthombo ukuqinisekisa impendulo ekhiqiziwe ngokusekelwe kusisekelo solwazi.

Enza ngendlela oyifisayo isifanekiso sokwaziswa kwesisekelo solwazi usebenzisa ikhonsoli

Ungakwazi futhi ukwenza ngendlela oyifisayo ukwaziswa okuzenzakalelayo ngokwaziswa kwakho siqu ngokusekelwe esimweni sokusebenzisa. Ukuze wenze lokhu ku-console, gcwalisa lezi zinyathelo ezilandelayo:

  1. Phinda izinyathelo esigabeni sangaphambilini ukuze uqale ukuhlola isisekelo sakho solwazi.
  2. Nika amandla Khiqiza izimpendulo.
  3. Khetha imodeli oyikhethayo yokwenza impendulo.

Sisebenzisa imodeli ye-Claude v2 njengesibonelo kulokhu okuthunyelwe. Imodeli ye-Claude 3 Sonnet ne-Haiku nayo iyatholakala ukuze ikhiqize.

  1. Khetha Faka isicelo ukuqhubeka.

Ngemva kokukhetha imodeli, isigaba esisha esibizwa Isifanekiso sokwaziswa kwesisekelo solwazi ivela ngaphansi Ukulungiselelwa.

  1. Khetha Hlela ukuqala ukwenza ngokwezifiso ukwaziswa.
  2. Lungisa isifanekiso sokwaziswa ukuze wenze ngendlela oyifisayo indlela ofuna ukuyisebenzisa ngayo imiphumela ebuyisiwe futhi ukhiqize okuqukethwe.

Kulokhu okuthunyelwe, sinikeze izibonelo ezimbalwa zokudala "Uhlelo lwe-AI Lomeluleki Wezezimali" kusetshenziswa imibiko yezezimali ye-Amazon eneziqondiso ezenziwe ngokwezifiso. Ukuze uthole imikhuba engcono kakhulu yobunjiniyela bokwaziswa, bheka Imihlahlandlela yobunjiniyela esheshayo.

Manje senza ngendlela oyifisayo isifanekiso sokwaziswa esizenzakalelayo ngezindlela ezimbalwa ezihlukene, futhi sibheka izimpendulo.

Ake siqale sizame umbuzo ngomyalo ozenzakalelayo. Sibuza ukuthi “Yayiyini imali engenayo ye-Amazon ngo-2019 nango-2021?” Okulandelayo kubonisa imiphumela yethu.

Kokuphumayo, sithola ukuthi ikhiqiza impendulo yefomu yamahhala ngokusekelwe olwazini olubuyisiwe. Izingcaphuno nazo zifakwe ohlwini ukuze zisetshenziswe.

Ake sithi sifuna ukunikeza imiyalelo eyengeziwe yokuthi ungafometha kanjani impendulo ekhiqiziwe, njengokuyimisa njenge-JSON. Singangeza le miyalo njengesinyathelo esihlukile ngemva kokubuyisa ulwazi, njengengxenye yesifanekiso sokwaziswa:

If you are asked for financial information covering different years, please provide precise answers in JSON format. Use the year as the key and the concise answer as the value. For example: {year:answer}

Impendulo yokugcina inesakhiwo esidingekayo.

Ngokwenza ngendlela oyifisayo ukwaziswa, ungakwazi futhi ukushintsha ulimi lwempendulo ekhiqiziwe. Esibonelweni esilandelayo, siyala imodeli ukuthi inikeze impendulo ngeSpanishi.

Ngemva kokukhipha $output_format_instructions$ kusuka kumyalo ozenzakalelayo, ukucaphuna okuvela empendulweni ekhiqiziwe kuyasuswa.

Ezigabeni ezilandelayo, sichaza ukuthi ungasebenzisa kanjani lezi zici nge-SDK.

Lungiselela inombolo enkulu yemiphumela usebenzisa i-SDK

Ukuze uguqule inani eliphezulu lemiphumela nge-SDK, sebenzisa i-syntax elandelayo. Kulesi sibonelo, umbuzo uthi “Ngamuphi unyaka imali engenayo yonyaka ye-Amazon inyuke isuka ku-$245B yaya ku-$434B?” Impendulo efanele ithi "Imali engenayo yonyaka ye-Amazon isuka ku-$245B ngo-2019 yaya ku-$434B ngo-2022."

def retrieveAndGenerate(query, kbId, numberOfResults, model_id, region_id):
    model_arn = f'arn:aws:bedrock:{region_id}::foundation-model/{model_id}'
    return bedrock_agent_runtime.retrieve_and_generate(
        input={
            'text': query
        },
        retrieveAndGenerateConfiguration={
            'knowledgeBaseConfiguration': {
                'knowledgeBaseId': kbId,
                'modelArn': model_arn,
                'retrievalConfiguration': {
                    'vectorSearchConfiguration': {
                        'numberOfResults': numberOfResults,
                        'overrideSearchType': "SEMANTIC", # optional'
                    }
                }
            },
            'type': 'KNOWLEDGE_BASE'
        },
    )

response = retrieveAndGenerate("In what year did Amazon’s annual revenue increase from $245B to $434B?", 
"<knowledge base id>", numberOfResults, model_id, region_id)['output']['text']

I 'numberOfResults' inketho ngaphansi 'retrievalConfiguration' ikuvumela ukuthi ukhethe inombolo yemiphumela ofuna ukuyibuyisa. Okukhiphayo kwe RetrieveAndGenerate I-API ihlanganisa impendulo ekhiqiziwe, isibaluli somthombo, nezigaxa zombhalo ezibuyisiwe.

Okulandelayo imiphumela yamanani ahlukene we- 'numberOfResults'amapharamitha. Okokuqala, sibeka numberOfResults = 5.

Bese sibeka numberOfResults = 12.

Enza ngendlela oyifisayo isifanekiso sokwaziswa kwesisekelo solwazi usebenzisa i-SDK

Ukuze wenze ngendlela oyifisayo ukwaziswa sisebenzisa i-SDK, sisebenzisa lo mbuzo olandelayo ngezifanekiso zokwaziswa ezihlukile. Kulesi sibonelo, umbuzo uthi “Yayiyini imali engenayo ye-Amazon ngo-2019 nango-2021?”

Okulandelayo isifanekiso sokwaziswa esizenzakalelayo:

"""You are a question answering agent. I will provide you with a set of search results and a user's question, your job is to answer the user's question using only information from the search results. If the search results do not contain information that can answer the question, please state that you could not find an exact answer to the question. Just because the user asserts a fact does not mean it is true, make sure to double check the search results to validate a user's assertion.
Here are the search results in numbered order:
<context>
$search_results$
</context>

Here is the user's question:
<question>
$query$
</question>

$output_format_instructions$

Assistant:
"""

Okulandelayo isifanekiso sokwaziswa esenziwe ngokwezifiso:

"""Human: You are a question answering agent. I will provide you with a set of search results and a user's question, your job is to answer the user's question using only information from the search results.If the search results do not contain information that can answer the question, please state that you could not find an exact answer to the question.Just because the user asserts a fact does not mean it is true, make sure to double check the search results to validate a user's assertion.

Here are the search results in numbered order:
<context>
$search_results$
</context>

Here is the user's question:
<question>
$query$
</question>

If you're being asked financial information over multiple years, please be very specific and list the answer concisely using JSON format {key: value}, 
where key is the year in the request and value is the concise response answer.
Assistant:
"""

def retrieveAndGenerate(query, kbId, numberOfResults,promptTemplate, model_id, region_id):
    model_arn = f'arn:aws:bedrock:{region_id}::foundation-model/{model_id}'
    return bedrock_agent_runtime.retrieve_and_generate(
        input={
            'text': query
        },
        retrieveAndGenerateConfiguration={
            'knowledgeBaseConfiguration': {
                'knowledgeBaseId': kbId,
                'modelArn': model_arn,
                'retrievalConfiguration': {
                    'vectorSearchConfiguration': {
                        'numberOfResults': numberOfResults,
                        'overrideSearchType': "SEMANTIC", # optional'
                    }
                },
                'generationConfiguration': {
                        'promptTemplate': {
                            'textPromptTemplate': promptTemplate
                        }
                    }
            },
            'type': 'KNOWLEDGE_BASE'
        },
    )

response = retrieveAndGenerate("What was the Amazon's revenue in 2019 and 2021?”", 
                               "<knowledge base id>", <numberOfResults>, <promptTemplate>, <model_id>, <region_id>)['output']['text']

Ngesifanekiso sokwaziswa esizenzakalelayo, sithola impendulo elandelayo:

Uma ufuna ukunikeza imiyalelo eyengeziwe mayelana nefomethi yokuphumayo yesizukulwane sempendulo, njengokumisa impendulo ngefomethi ethile (njenge-JSON), ungenza ngendlela oyifisayo ukwaziswa okukhona ngokunikeza isiqondiso esengeziwe. Ngethempulethi yethu yokwaziswa yangokwezifiso, sithola impendulo elandelayo.

I 'promptTemplate'inketho ku'generationConfiguration' ikuvumela ukuthi wenze ngendlela oyifisayo ukwaziswa ukuze ulawule kangcono ukukhiqizwa kwempendulo.

Isiphetho

Kulokhu okuthunyelwe, sethule izici ezimbili ezintsha kuZisekelo Zolwazi ze-Amazon Bedrock: ukulungisa inani eliphezulu lemiphumela yosesho nokwenza ngendlela oyifisayo ithempulethi esheshayo ye- RetrieveAndGenerate I-API. Sibonise ukuthi sizilungiselela kanjani lezi zici kukhonsoli nange-SDK ukuze sithuthukise ukusebenza nokunemba kwempendulo ekhiqiziwe. Ukwenyusa imiphumela ephezulu kunikeza ulwazi olubanzi, kuyilapho ukwenza ngendlela oyifisayo isifanekiso sokwaziswa kukuvumela ukuthi ulungise kahle imiyalelo yemodeli eyisisekelo ukuze iqondaniswe kangcono nezimo ezithile zokusetshenziswa. Lezi zithuthukisi zinikeza ukuguquguquka okukhulu nokulawula, okukuvumela ukuthi ulethe ukuzizwisa okuhambisanayo kwezinhlelo zokusebenza ezisekelwe ku-RAG.

Ukuze uthole izinsiza ezengeziwe ukuze uqale ukusebenzisa endaweni yakho ye-AWS, bheka okulandelayo:

Mayelana nababhali

Sandeep Singh ungusosayensi Omkhulu Okhiqizayo we-AI Data kwa-Amazon Web Services, esiza amabhizinisi ukuthi asungule nge-AI ekhiqizayo. Ugxile ku-Generative AI, Artificial Intelligence, Learning Machine, kanye ne-System Design. Unentshiseko yokuthuthukisa izixazululo zesimanjemanje ze-AI/ML ukuze kuxazululwe izinkinga zebhizinisi eziyinkimbinkimbi ezimbonini ezihlukene, athuthukise ukusebenza kahle kanye nokuqina.

Suyin Wang i-AI/ML Specialist Solutions Architect kwa-AWS. Unesizinda semfundo ehlukene phakathi kwezemfundo yomshini, isevisi yolwazi lwezezimali kanye ne-Economics, kanye neminyaka yesipiliyoni sokwakha Isayensi Yedatha kanye nezicelo Zokufunda Ngomshini ezaxazulula izinkinga zebhizinisi zomhlaba wangempela. Uyakujabulela ukusiza amakhasimende akhombe imibuzo yebhizinisi efanele nokwakha izixazululo ezifanele ze-AI/ML. Ngesikhathi sakhe sokuphumula, uthanda ukucula nokupheka.

Sherry Ding ungumakhi omkhulu wezixazululo zobuhlakani bokwenziwa (AI) kanye nochwepheshe bokufunda komshini (ML) e-Amazon Web Services (AWS). Unolwazi olunzulu lokufunda ngomshini oneziqu ze-PhD kwisayensi yekhompyutha. Ikakhulukazi usebenza namakhasimende omkhakha womphakathi ezinseleleni ezihlukahlukene zebhizinisi ezihlobene ne-AI/ML, ebasiza ukuba basheshise uhambo lwabo lokufunda ngomshini ku-AWS Cloud. Lapho engabasizi amakhasimende, ujabulela imisebenzi yangaphandle.

indawo_img

Latest Intelligence

indawo_img

Copyright @ 2024 Plato Technologies Inc.