Thumela amamodeli we-GPT-OSS nge-Amazon Bedrock Custom Model Templeveli

0 1 7 minutes read

Thumela amamodeli we-GPT-OSS nge-Amazon Bedrock Custom Model Templeveli

I-Amazon Bedrock Custom Model Section Manje isekela amamodeli we-Openai ngezinsimbi ezivulekile, kufaka phakathi ukuhlukahluka kwe-GPT-OSS ngamapharamitha ayizigidi eziyizinkulungwane ezingama-120. Amamodeli we-GPT-OSS anikeza amakhono okubonisana futhi angasetshenziswa nge-Opelai Chat Comptions API. Ngokugcina ukuhambisana okugcwele kwe-OpenAi API, izinhlangano zingahambisa izinhlelo zazo ezikhona ku-AWS, zithola ukuphepha kwebanga lebhizinisi, ukukala, nokulawulwa kwezindleko.

Kulokhu okuthunyelwe, sibonisa ukuthi ungathumela kanjani imodeli ye-GPT-OSS-20B-20B e-Amazon Bedrock usebenzisa imodeli yangokwezifiso ngenkathi kugcinwa ukuhambisana okuphelele kwe-API nezinhlelo zakho zamanje.

Sibutsetelo se-Amazon Bedrock Custom Model Model

I-Amazon Bedrock Custom Model Lectole ikuvumela ukuthi ulethe amamodeli ahleliwe kwimvelo efanayo yeseva lapho ufinyelela khona amamodeli wesisekelo (FMS). Uthola i-API eyodwa enobunye ngakho konke; Awudingi ukujikijela ama-Endpoindles amaningi noma uphathe ingqalasizinda ehlukile.

Ukuze usebenzise lesi sici, layisha amafayela wakho wemodeli ku-Amazon Isitoreji Service (i-Amazon S3), bese uqala ukungenisa nge-Amazon Bedrock Console. Ama-AWS aphatha ukuphakanyiswa okusindayo, kufaka phakathi i-GPU ekuhlinzekeni, ukumisa amaseva wokuphakanyiswa, nokulinganisa ngokuzenzakalela ngokuya ngesidingo. Ungagxila kwizicelo zakho ngenkathi ama-AWS ephatha ingqalasizinda.

Amamodeli we-GPT-OSS asekela i-Opelai Chat Compling API, kufaka phakathi ama-Armys agcinwe ngemiyalezo, izincazelo zomsebenzisi (uhlelo, umsebenzisi, noma umsizi), kanye nezakhiwo ezijwayelekile zokuphendula ngamamethrikhi wethokheni. Ungakhomba izinhlelo zakho zokusebenza kuma-amazon bedrock endpoints, futhi zizosebenza nezinguquko ezincane kwikhodi yakho.

Ukubuka konke amamodeli we-GPT-OSS

Amamodeli we-GPT-OSS angamamodeli wokuqala wesisindo esivulekile esingesisindo esivulekile se-GPT-2, ekhishwe ngaphansi kwelayisense le-Apache 2.0. Ungalanda, ushintshe, futhi uzisebenzise ngaphandle kwezindleko ezengeziwe, kufaka phakathi izinhlelo zokusebenza zezentengiso. Lawa mamodeli agxile ekubonisaneni, ukusetshenziswa kwamathuluzi, nokuthunyelwa okusebenzayo.Khetha imodeli efanele yezidingo zakho:

I-GPT-OSS-20B (amapharamitha ayizigidi eziyizinkulungwane ezingama-21) – Le modeli ilungele izinhlelo zokusebenza lapho ijubane nokusebenza kahle kakhulu. Ngaphandle kokuba namapharamitha ayizigidi eziyizinkulungwane ezingama-21, kusebenze kuphela ama-3.6 billion ngethokheni ngalinye, ngakho-ke isebenza kahle kumadivayisi we-Edge ngememori eyi-16 kuphela. Ngezendlalelo ezingama-24, ochwepheshe abangama-32 (ama-4 asebenzayo ngakunye), kanye newindows 128K percomis, kufana nokusebenza kwe-O3-mini ye-O3Win ngenkathi ukwazi ukufaka endaweni yangakini ngezikhathi zokuphendula ngokushesha.
I-GPT-OSS-120B (amapharamitha ayizigidi eziyizinkulungwane ezingama-117) – Yakhelwe imisebenzi eyinkimbinkimbi yokubonisana efana namakhodi, izibalo, kanye ne-agentic ithuluzi sebenzisa, kusebenze amapharamitha ayizigidi eziyizinkulungwane ezingama-5.1 ngethokheni ngalinye. Ngezendlalelo ezingama-36, ochwepheshe abangu-128 (abangu-4 abasebenzayo ngakunye), kanye newindows 128k elithile, lihambisana nokusebenza kwe-O4-mini e-OpenAi ngenkathi lisebenza kahle ku-80GB GPU.

Womabili amamodeli asebenzisa ochwepheshe bezinhlanganisela (ama-moe) wochwepheshe wezakhiwo zemodeli (ochwepheshe) baphatha izinhlobo ezahlukene zemisebenzi, kuphela ochwepheshe abafanele kakhulu abasebenze ngesicelo ngasinye. Le ndlela ikunikeza ukusebenza okunamandla ngenkathi ugcina izindleko ze-computational zilawuleka.

Ukuqonda ifomethi yemodeli ye-GPT-OSS

Lapho ulanda amamodeli we-GPT-OSS kusuka ekugwingweni kobuso, uthola izinhlobo eziningana zefayela ezisebenza ngokubambisana:

Amafayela Wesisindo (.Safetensisors) – Amapharamitha wangempela wemodeli
Amafayela wokucushwa (i-Config.json) – Izilungiselelo ezichaza ukuthi imodeli isebenza kanjani
Amafayela we-Tokenizer – Hambisa ukucubungula umbhalo
I-Index File (Model.Safetensisors.index.json) – Idatha yesisindo yamamephu kumafayela athile

Ifayela le-Index lidinga ukwakheka okuthile ukuze lisebenze ne-Amazon Bedrock. Kumele ifake a metadata insimu ngezinga lempande. Lokhu kungaba yize ({}) Noma uqukethe usayizi wemodeli ophelele (okumele ube ngaphansi kwama-200 GB ngamamodeli wombhalo).

Amamodeli avela ekugwileni ngobuso kwesinye isikhathi afaka amasimu we-metadata angeziwe afana total_parameters ukuthi i-Amazon Bedrock ayisekeli. Kufanele ususe lokhu ngaphambi kokungenisa. Isakhiwo esifanele kufanele sibukeke njengekhodi elandelayo:

{
"metadata": {},
"weight_map": {
"lm_head.weight": "model-00009-of-00009.safetensors",
    ...
  }
}

Qiniseka ukuthi awufaki metal umkhombandlela ngaphambi kokulayisha i-Amazon S3 Layisha.

Ukubuka konke

Kulokhu okuthunyelwe, sihamba ngenqubo ephelele yokuthumela esebenzisa i-Amazon Bedrock Custom Model Model. Sisebenzisa imodeli ye-tonic / med-GPT-OSS-20B-20B, inguqulo ehlelwe kahle ye-Opelai-OSS-20B elungiselelwe ngokuqondile ukucabanga kwezokwelapha kanye nemiyalo elandelayo.

Inqubo yokuhambisa ukuthunyelwa ifaka izinyathelo ezine eziphambili:

Landa amafayela wemodeli kusuka ekuggqeni kobuso bese ulungiselela ama-AWS.
Faka amafayela wemodeli ku-Amazon S3.
Ngenisa usebenzisa i-Amazon Bedrock Custom Model Leccel ukuletha imodeli yakho e-Amazon Bedrock.
Nxusa imodeli yakho ngezingcingo ezihambisanayo ze-API ezihambisanayo ukuze uvivinye ukuthunyelwa kwakho.

Umdwebo olandelayo uveza ukuhamba komsebenzi wokuhanjiswa.

Izimfuneko

Ngaphambi kokuthi uqale ukuthumela imodeli yakho ye-GPT-OSS, qiniseka ukuthi unakho okulandelayo:

I-akhawunti ye-AWS esebenzayo enezimvume ezifanele
Ubunikazi be-AWS ubunikazi kanye nokuphathwa kokufinyelela (iam) ku:
- Dala Imisebenzi Yokungenisa Imodeli e-Amazon Bedrock
- Faka amafayela kuma-amazon s3
- Inxusa amamodeli ngemuva kokuthunyelwa
- Sebenzisa Indima Yesevisi Yezemisebenzi Yangokwezifiso
Ibhakede le-S3 kwisifunda sakho se-AWS
Cishe i-40 GB yesikhala sediski sendawo sokulanda imodeli
Ukufinyelela esifundeni sase-US East 1 (N. Virginia) (kudingekile kumamodeli we-GPT-OSS asekelwe ngokwezifiso)
I-AWS Command Line Interface (AWS CLI) I-2.X efakiwe
Ukugibela ubuso be-CLI (faka nge pip install -U "huggingface_hub[cli]"Isihlehlukene

Landa futhi ulungiselele amafayela wemodeli

Ukulanda imodeli ye-GPT-OSS usebenzisa umtapo wezincwadi we-hug ebusweni ngokudluliselwa okusheshayo okunikwe amandla, sebenzisa ikhodi elandelayo:

import os
os.environ['HF_HUB_ENABLE_HF_TRANSFER'] = '1'
from huggingface_hub import snapshot_download

local_dir = snapshot_download(
    repo_id="Tonic/med-gpt-oss-20b",
    local_dir="./med-gpt-oss-20b",
)print(f"Download complete! Model saved to: {local_dir}")

Ngemuva kokuthi ukulanda kuqediwe (imizuzu engu-10-20 ye-40 GB), qinisekisa model.safetensors.index.json Isakhiwo sefayela. Ukuhlela uma kudingeka ukuqiniseka ukuthi metadata Inkambu ikhona (ingabi nalutho):

{
"metadata": {},
"weight_map": {
"lm_head.weight": "model-00009-of-00009.safetensors",
    ...
  }
}

ec2-user@ip-XYZ  ~/gptoss/med-gpt-oss-20b  ls  -lah
total 39G
drwxr-xr-x. 3 ec2-user ec2-user  16K Nov 10 19:38 .
drwxr-xr-x. 3 ec2-user ec2-user   44 Nov 10 21:31 ..
drwxr-xr-x. 3 ec2-user ec2-user   25 Nov 10 18:57 .cache
-rw-r--r--. 1 ec2-user ec2-user  17K Nov 10 18:57 chat_template.jinja
-rw-r--r--. 1 ec2-user ec2-user 1.6K Nov 10 18:57 config.json
-rw-r--r--. 1 ec2-user ec2-user  160 Nov 10 18:57 generation_config.json
-rw-r--r--. 1 ec2-user ec2-user 1.6K Nov 10 18:57 .gitattributes
-rw-r--r--. 1 ec2-user ec2-user 4.2G Nov 10 18:57 model-00001-of-00009.safetensors
-rw-r--r--. 1 ec2-user ec2-user 4.6G Nov 10 18:57 model-00002-of-00009.safetensors
-rw-r--r--. 1 ec2-user ec2-user 4.6G Nov 10 18:57 model-00003-of-00009.safetensors
-rw-r--r--. 1 ec2-user ec2-user 4.6G Nov 10 18:57 model-00004-of-00009.safetensors
-rw-r--r--. 1 ec2-user ec2-user 4.6G Nov 10 18:57 model-00005-of-00009.safetensors
-rw-r--r--. 1 ec2-user ec2-user 4.6G Nov 10 18:57 model-00006-of-00009.safetensors
-rw-r--r--. 1 ec2-user ec2-user 4.6G Nov 10 18:57 model-00007-of-00009.safetensors
-rw-r--r--. 1 ec2-user ec2-user 4.6G Nov 10 18:57 model-00008-of-00009.safetensors
-rw-r--r--. 1 ec2-user ec2-user 2.6G Nov 10 18:57 model-00009-of-00009.safetensors
-rw-r--r--. 1 ec2-user ec2-user  33K Nov 10 19:38 model.safetensors.index.json
-rw-r--r--. 1 ec2-user ec2-user 5.4K Nov 10 18:57 README.md
-rw-r--r--. 1 ec2-user ec2-user  440 Nov 10 18:57 special_tokens_map.json
-rw-r--r--. 1 ec2-user ec2-user 4.2K Nov 10 18:57 tokenizer_config.json
-rw-r--r--. 1 ec2-user ec2-user  27M Nov 10 18:57 tokenizer.json

Faka amafayela wemodeli ku-Amazon S3

Ngaphambi kokuthi ungenise imodeli yakho, kufanele ugcine amafayela wemodeli kubhakede le-S3 lapho i-Amazon Bedrock ingazithola khona. Qedela lezi zinyathelo ezilandelayo:

Ku-Amazon S3 Console, khetha Ibhakede kufasitelana lokuzulazula.
Dala ibhakede elisha noma uvule eyodwa ekhona.
Faka amafayela wakho wemodeli.

Ngenye indlela, layisha amafayela ebhakedeni le-S3 esifundeni sakho se-Amazon Bedrock usebenzisa i-AWS CLI:

aws s3 sync ./med-gpt-oss-20b/ s3://amzn-s3-demo-bucket/med-gpt-oss-20b/

Ukulayishwa kwe-40 GB ngokuvamile kuqedelwa ngemizuzu engu-5 ukuya kwe-10. Qinisekisa amafayela ayalayishwa:

aws s3 ls s3://amzn-s3-demo-bucket/med-gpt-oss-20b/ --human-readable

Isithonjana esilandelayo sibonisa isibonelo samafayela kubhakede le-S3.

Qaphela i-S3 URI yakho (ngokwesibonelo, s3://amzn-s3-demo-bucket/med-gpt-oss-20b/) Ukusebenzisa umsebenzi wokungenisa.

Amafayela okukhipha abethelwe ngokucushwa kwe-Encryption yebhakede le-S3. Lokhu kubhalwe ngemfihlo kungaba nge-SSE-S3 Server-Side Service noma nge-AWS Key Management Service (AWS KMS) Imfihlo ye-SSE-KMS, kuya ngokuthi ubeka kanjani i-S3 Bucket.

Imodeli yokungenisa usebenzisa ukungena kwe-Amazon Bedrock ngokwezifiso

Manje njengoba amafayela wakho wemodeli alayishwe ku-Amazon S3, ungangenisa imodeli e-Amazon Bedrock, lapho izocutshungulwa khona futhi yenziwa itholakale ukuze itholakale ukuze itholakale ukuze itholakale ukuze itholakale. Qedela lezi zinyathelo ezilandelayo:

E-Amazon Bedrock Console, khetha Amamodeli angenisiwe kufasitelana lokuzulazula.
Qoka Imodeli Yengenisa.
Ingomane Igama lemodelifaka gpt-oss-20b.
Ingomane Umthombo wokungenisa amamodelikhetha I-Amazon S3 Bucket.
Ingomane Indawo ye-S3faka s3://amzn-s3-demo-bucket/med-gpt-oss-20b/.
Ingomane Ukufinyelela kwensizakalokhetha Dala futhi usebenzise indima entsha yensizakalo. I-Amazon Bedrock Console ikhiqiza ngokuzenzakalelayo indima enobudlelwano obufanele bokuthembela kanye ne-Amazon S3 Read Reintsholotions.
Qoka Ngenisa ezweni isifanekiso ukuqala umsebenzi wokungenisa.

Ukuze usebenzise i-AWS CLI, sebenzisa ikhodi elandelayo:

aws bedrock create-model-import-job 
  --job-name "gpt-oss-20b-import-$(date +%Y%m%d-%H%M%S)" 
  --imported-model-name "gpt-oss-20b" 
  --role-arn "arn:aws:iam::YOUR-ACCOUNT-ID:role/YOUR-ROLE-NAME" 
  --model-data-source "s3DataSource={s3Uri=s3://amzn-s3-demo-bucket/med-gpt-oss-20b/}"

Ukungenisa imodeli ngokuvamile kuqedela ngemizuzu engu-10-16 yemodeli yepharamitha engama-20b. Ungabheka inqubekelaphambili e-Amazon Bedrock Console noma usebenzisa i-AWS CLI. Lapho usuqedile, phawula eyakho importedModelArnosebenzisa ukunxusa imodeli.

Imodeli ye-API ehambisanayo

Ngemuva kokuthi ukungeniswa kwemodeli yakho sekuqediwe, ungayivivinya usebenzisa i-OpenAi Chat Compling Comptions formation ukuqinisekisa ukuthi iyasebenza njengoba kulindeleke:

Dala ifayela eliqanjwe test-request.json Ngokuqukethwe okulandelayo:

{
  "messages": [
    {
      "role": "system",
      "content": "You are a helpful AI assistant."
    },
    {
      "role": "user",
      "content": "What are the common symptoms of Type 2 Diabetes?"
    }
  ],
  "max_tokens": 500,
  "temperature": 0.7
}

Sebenzisa ama-AWS CLI ukuthumela isicelo kwisicelo sakho semodeli engenisiwe:

aws bedrock-runtime invoke-model 
  --model-id "arn:aws:bedrock:us-east-1:YOUR-ACCOUNT-ID:imported-model/MODEL-ID" 
  --body file://test-request.json 
  --cli-binary-format raw-in-base64-out 
  response.json
cat response.json | jq '.'

Impendulo ibuya ngefomethi evamile ye-OpenAi:

{
  "id": "chatcmpl-f06adcc78daa49ce9dd2c58f616bad0c",
  "object": "chat.completion",
  "created": 1762807959,
  "model": "YOUR-ACCOUNT-ID-MODEL-ID",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Type 2 Diabetes often presents with a range of symptoms...",
        "refusal": null,
        "function_call": null,
        "tool_calls": []
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 98,
    "completion_tokens": 499,
    "total_tokens": 597
  }
}

Isakhiwo Sokuphendula sifana nefomethi ka-Opelai impela-choices Kuqukethe impendulo, usage ihlinzeka ukubalwa kwamathokheni, futhi finish_reason kubonisa isimo sokuqeda. Ikhodi ekhona yokuphendula evulekile ye-OpenAI isebenza ngaphandle kokuguqulwa.

Inzuzo enamandla yale modeli ukucaca kwayo. Le khasi reasoning_content Inkambu isinikeza imininingwane ngenqubo yangaphakathi yokubonisana yangaphakathi ngaphambi kokuthi ikhiphe impendulo yokugcina. Leli zinga lokusobala alitholakali nge-API evaliwe yomthombo.

Hlanza

Uma usuqedile, hlanza izinsizakusebenza zakho ukugwema izindleko ezingadingekile:

aws bedrock delete-imported-model --model-identifier "arn:aws:bedrock:us-east-1:ACCOUNT:imported-model/MODEL-ID"

aws s3 rm s3://amzn-s3-demo-bucket/med-gpt-oss-20b/ --recursive

Uma ungasadingi indima ye-IAM, susa usebenzisa i-IAM Console.

Athuke kusuka ku-accaiya uye e-amazon bedrock

Ukufuduka kusuka ku-OpenAI kudinga ushintsho oluncane lwekhodi-kuphela indlela yokuvuselela lapho izinhlaka zemiyalezo zihlala zifana.

Nge-Opena, sebenzisa ikhodi elandelayo:

import openai
response = openai.ChatCompletion.create(model="....", messages=[...])

Okwe-Amazon Bedrock, sebenzisa ikhodi elandelayo:

import boto3, json
bedrock = boto3.client('bedrock-runtime')
response = bedrock.invoke_model(
    modelId='arn:aws:bedrock:us-east-1:ACCOUNT:imported-model/MODEL-ID',
    body=json.dumps({"messages": [...]})
)

Ukufuduka kuqondile, futhi uzothola izindleko zokubikezela, ubumfihlo bedatha engcono, kanye nekhono amamodeli amahle ngezidingo zakho ezithile.

Imikhuba emihle kakhulu

Cabanga ngemikhuba emihle elandelayo:

Ukuqinisekiswa kwefayela – ngaphambi kokulayisha, qinisekisa model.safetensors.index.json Inesakhiwo se-metadata esifanele, amafayela we-SafeTenses abhalwe phansi akhona, futhi ama-tokenizers asekelwa. Ukuqinisekiswa kwasendaweni kusindisa isikhathi sokuzama kabusha.
Ukuvikeleka – E-Amazon Bedrock Console, dala izindima ze-IAM ngokuzenzakalela ezinemvume enobukhulu. Kumamodeli amaningi, sebenzisa iziqalo ze-S3 ezihlukile ukuze zilondoloze ukwahlukaniswa.
Ukuhumusho – Sebenzisa izindlela ezichazayo ze-S3 (gpt-oss-20b-v1.0/) noma ukungeniswa kwamagama okungenisa umsebenzi wokulandela umkhondo.

Ukubeka impahla ephakeme

Ukhokhiswa ngokuhambisana nokuhambisana namamodeli wangokwezifiso owungenisa e-Amazon Bedrock. Ngemininingwane engaphezulu, bheka ukubala izindleko zokusebenzisa imodeli yangokwezifiso ne-Amazon Bedrock Pricing.

Ukugcina

Kulokhu okuthunyelwe, sikhombise ukuthi singayithumela kanjani amamodeli we-GPT-OSS e-Amazon Bedrock esebenzisa imodeli yangokwezifiso ngenkathi kugcinwa ukuhambisana okugcwele kwe-OpenAi API. Manje usungahambisa izinhlelo zakho ezikhona ku-AWS ngokushintsha kwekhodi encane futhi uthole izinzuzo zebhizinisi, kufaka phakathi ukulawulwa okuphelele kwemodeli, amakhono okuqagela okuhle, izintengo zokuqagela, ubumfihlo bedatha ethuthukisiwe.

Ukulungele ukuqala? Nazi izinyathelo zakho ezilandelayo:

Khetha usayizi wakho wemodeli – Qala ngemodeli ye-20B yezimpendulo ezisheshayo, noma usebenzise i-120B ehlukile yemisebenzi yokubonisana eyinkimbinkimbi
Setha imvelo yakho – Qinisekisa ukuthi unezimvume ze-AWS ezidingekayo kanye ne-S3 Bucket Access
Zama i-Amazon Bedrock Console – Ngenisa imodeli yakho yokuqala ye-GPT-OSS usebenzisa i-Amazon Bedrock Console
Hlola izici ezithuthukile – Cabanga ngokuhleleka okuhle ngedatha yakho yokuphathelene ngemuva kokusetha kwakho okuyisisekelo kuyasebenza

I-Amazon Bedrock Custom Model Lectolovel iyatholakala ezifundeni eziningi, ngokusekelwa ukunwebeka ezifundeni ezingeziwe maduze. Bheka ukusekelwa kwesikhashana yi-AWS Region e-Amazon Bedrock yezibuyekezo zakamuva. Amamodeli we-GPT-OSS ekuqaleni azotholakala esifundeni sase-US-East-1 (N. Virginia).

Unemibuzo noma impendulo? Xhuma nethimba lethu ngokusebenzisa ama-AWS Re: Okuthunyelwe kwe-Amazon Bedrock-Singathanda ukuzwa ngesipiliyoni sakho.

Mayelana nababhali

Prashant Patel Unjiniyela wokuthuthukisa isoftware aphezulu e-AWS Bedrock. Unothando ngokulinganisa amamodeli amakhulu wezilimi ngezinhlelo zokusebenza zebhizinisi. Ngaphambi kokujoyina ama-AWS, wasebenza e-IBM ekukhiqizeni imithwalo emikhulu ye-AI / ML kuma-Kubernete. I-Prashant ineziqu ze-master kusuka ku-Nyu Tandon School of Engineering. Yize kungasebenzi, uyakujabulela ukuhamba futhi edlala nezinja zakhe.

Ainesh sootha ngunjiniyela wokuthuthukisa isoftware ngama-AWS. Unothando ngokusebenza kahle kokusebenza kanye nokulinganisa amamodeli amakhulu wezilimi ngezinhlelo zokusebenza zebhizinisi. Ngaphambi kokujoyina i-Aw Bedrock, wasebenza ezinhlelweni zokufakazela ubuqiniso e-Amazon's Vele Watch Out Technology. I-Ainesh ine-degree ye-bachelor enjiniyela bekhompyutha kusuka ePurdue University. Yize kungasebenzi, uyakujabulela ukudlala isiginci kanye nezincwadi zokufunda.

Sandeep akhouri ngumkhiqizo onolwazi kanye nomholi we-Go-To-Market (GTM) oneminyaka engaphezu kwengu-20 yesipiliyoni ekuphathweni komkhiqizo, ubunjiniyela kanye nasemakethe. Ngaphambi kokuba neqhaza lakhe elikhona njengamanje, imikhiqizo yokuphathwa komkhiqizo ye-SALEEEEP I-LED ezinkampanini zezobuchwepheshe eziholayo, kufaka phakathi ama-splonk, ama-KX System, HazelCast kanye ne-Software AG. Unentshisekelo nge-agentic AI, imodeli ngokwezifiso nokushayela umthelela wangempela womhlaba nge-ai generative ai.