矢量数据库的兴起.pptx

《矢量数据库的兴起.pptx》由会员分享，可在线阅读，更多相关《矢量数据库的兴起.pptx（17页珍藏版）》请在三个皮匠报告上搜索。

1、Kevin PetrieVP ResearchMay 8,2024,Rise of the Vector DatabaseEnabling Generative AI,Generative AI Adoption Stages,Most GenAI adopters are integrating language models into multi-faceted workflows,1.Language Model PlatformsEmployees use LLM tools as standalone platformsOpenAI ChatGPTGoogle BardHugging

2、 Face BLOOM,2.LM Assistants within ToolsEmployees use LLM functions within commercial toolsSalesforce EinsteinGitHub CopilotSAP Joule,3.LM-Driven WorkflowsCompanies build LM functions into multi-faceted workflowsCustom tools,applications,and integrations,PARAMETER COUNTS,DOMAIN-SPECIFIC DATA,TODAYS

3、FOCUS,Vector databases play a critical role in all three architectural approaches,+,-,Architectural Approaches to Domain-Specific Language Models,MOST COMMON,Approaches vary by industry,maturity,and domain-specific requirements,Adoption Patterns,MOST COMMON,VECTORSParameters that are embedded into v

4、ector databases,Language models are trained on a corpus of text to generate strings of words,TOKENS“When,”“in,”“the”,CHUNKSPreamble,Body,Conclusion,PARAMETERS Numbers that measure relationships of chunks and tokens,ATTENTION NETWORKUses parameters to weight the influence of tokens,LM INFERENCEUses a

5、ttention network to generate text when prompted,EXAMPLE SOURCE:DECLARATION OF INDEPENDENCE,Generative AI Needs New Models and Structures,“When in the Course of human events”,Vector databases store,organize,maintain,and deliver high-dimension data,Captures high-dimension data that has many features p

6、er data point:words,images,etc.Indexes vectors based on similarity;governs access and usagePerforms DML operations and semantic searches,Enter the Vector Database,Vector databases rely on an embedding model to index data points according to their similarity to one another,Prashanth Rao,The Data Quar