Inkundla yokuhlaziya I-Databricks yethule imodeli yolimi oluyisisekelo yomthombo ovulekile, ngethemba lokuthi amabhizinisi azokhetha ukusebenzisa amathuluzi ayo ukuze agxumele kubhendi ye-LLM.

I-biz, eyasungulwa eduze kwe-Apache Spark, ishicilele inqwaba yamabhentshimakhi abathi i-LLM yenhloso yayo ejwayelekile - ebizwa ngokuthi i-DBRX - yehlula izimbangi zomthombo ovulekile ekuqondeni kolimi, ukuhlela, kanye nezibalo. Umthuthukisi uphinde wathi yehlula i-OpenAI's proprietary GPT-3.5 kuzo zonke izindlela ezifanayo.

I-DBRX yasungulwa ngabakwa-Mosaic AI, okuthi Databricks etholiwe ngamadola ayizigidi eziyizinkulungwane ezingu-1.3, futhi waqeqeshwa ku-Nvidia DGX Cloud. I-Databricks ithi ithuthukise i-DBRX ukuze isebenze kahle ngalokho ekubiza nge-mix-of-expert architecture (MoE) - lapho amanethiwekhi ochwepheshe abaningi noma abafundi behlukanisa inkinga.

I-Databricks ichaze ukuthi imodeli inamapharamitha ayizigidi eziyizinkulungwane ezingama-132, kodwa yizigidigidi ezingama-36 kuphela ezisebenzayo kunoma yikuphi okokufaka okukodwa.

UJoel Minnick, iphini likamongameli wezokumaketha weDatabricks, utshele IRejista: “Leso yisizathu esikhulu esenza imodeli ikwazi ukugijima ngendlela esebenza ngayo, kodwa futhi igijima ngesivinini esimangalisayo. Ngokwezinto ezibonakalayo, uma usebenzisa noma yiluphi uhlobo lwama-chatbots amakhulu akhona namuhla, cishe ujwayele ukulinda futhi ubukele impendulo ikhiqizwa. Nge-DBRX iseduze ngokuphazima kweso.”

Kodwa ukusebenza kwemodeli ngokwayo akulona iphuzu le-Databricks. I-biz, phela, yenza i-DBRX itholakale khulula ku-GitHub futhi Ubuso Obumbambayo.

I-Databricks inethemba lokuthi amakhasimende asebenzisa imodeli njengesisekelo sama-LLM awo. Uma lokho kwenzeka kungase kuthuthukise ama-chatbots ekhasimende noma ukuphendula imibuzo yangaphakathi, kuyilapho kubonisa ukuthi i-DBRX yakhiwa kanjani kusetshenziswa amathuluzi obunikazi be-Databricks.

I-Databricks ihlanganisa idathasethi lapho i-DBRX yathuthukiswa khona kusetshenziswa amanothibhuku e-Apache Spark ne-Databricks ukuze kucutshungulwe idatha, Ikhathalogi Yobumbano yokuphathwa kwedatha nokubusa, kanye ne-MLflow yokulandelela ukuhlolwa.

UMinnick uveze ukuthi ukutshalwa kwezimali kwebhizinisi kuma-LLM kwabambezeleka ngenxa yokwesaba ubunikazi nokuphatha okuvela eceleni. "Kudingeka uhambise idatha kubantu besithathu, ukungabi nobunikazi phezu kwezisindo zemodeli, ukungakwazi ukulawula ngokugcwele ukubuswa kwedatha ekupheleni kokuphela - lezi yizinto ezibanciphisa," echaza.

"Esizimisele ukukwakha bekuyimodeli esebenza kahle kakhulu ... amabhizinisi angayisebenzisa ukuze alethe izinhlelo zawo zokusebenzisa amacala athile."

U-Hyoun Park, i-CEO kanye nomhlaziyi omkhulu kwa-Amalgam Insights, ukubonile ukubaluleka kwe-DBRX ukuthi i-Databricks ingabonisa ukuthi imodeli yakhiwe kanjani, isinyathelo ngesinyathelo, njengenqubo yokuthi amanye amabhizinisi ayilandele futhi ashune kahle.

"Leyo nhlanganisela yohlu lozalo, ukubonakala, ukuphindaphinda, kanye nobunikazi bemodeli ekuhleleni imodeli ekupheleni kuya kokuphela, ukuhlolwa, nokusebenza kubalulekile."

UPark waphawula ukuthi uyaqonda ukuthi i-Databricks yayakhe amamodeli ngokwezifiso angaphezu kuka-50,000 kumakhasimende kakade. “Yileyo nhlanganisela yesipiliyoni sokwakha imodeli nekhono lokukwenza ngesilinganiso esisezingeni eliphezulu ngokuhambisana nemizamo engcono kakhulu yangasese nevulekile eyenza lesi simemezelo siphawuleke kimi ngokombono we-IT webhizinisi.”

Izindaba ze-DBRX zidlala ngokumelene nesizinda esishintshayo sokuncintisana seDathabricks. I-biz inobambiswano lwesu lwesikhathi eside ne-Microsoft, okuholele ku-Azure Databricks - lapho abasebenzisi bathenjiswa khona izinsizakalo zedatha ezihlanganisiwe ezisondelene nepulatifomu yamafu ye-Redmond giant.

Kodwa kusukela ukunikezwa kwethulwa ngo-2017, iMicrosoft isithuthele emakethe ye-lakehouse ye-Databricks - lapho abasebenzisi banikezwa khona ukugcinwa kwedatha kanye namachibi edatha endaweni eyodwa - futhi ithembisa abasebenzisi ama-LLM asezingeni lebhizinisi nge yayo $10 billion Ubambiswano lwe-OpenAI. Endaweni yayo Yendwangu, I-Microsoft inganikeza futhi "isibuko" kusuka ezinhlelweni zayo zedatha yedatha i-Azure Cosmos DB ne-Azure SQL DB, enikeza ukufinyelela ezinsizeni zokuhlaziya ngaphandle kokuhambisa idatha.

Umbuzo ovulekile olenga phezu kwamasu akho kokubili i-Databricks kanye ne-Microsoft yilapho kuzofika uzamcolo wokutshalwa kwezimali okulindelekile kubuchwepheshe be-LLM. NgoJanuwari, Kubikezela uGartner ukusetshenziswa kwebhizinisi kubuchwepheshe ngeke kuze kulo nyaka, futhi kuzoba nomthelela omncane kokunye ukutshalwa kwezimali kwe-IT. ®

• I-SEO Powered Content & PR Distribution. Khuliswa Namuhla.
• I-PlatoData.Network Vertical Generative Ai. Zinike Amandla. Finyelela Lapha.
• I-PlatoAiStream. I-Web3 Intelligence. Ulwazi Lukhulisiwe. Finyelela Lapha.
• I-PlatoESG. Ikhabhoni, I-CleanTech, Amandla, Environment, Ilanga, Ukuphathwa Kwemfucuza. Finyelela Lapha.
• I-PlatoHealth. I-Biotech kanye ne-Clinical Trials Intelligence. Finyelela Lapha.
• Source: https://go.theregister.com/feed/www.theregister.com/2024/03/28/databricks_dbrx_llm/