2769 - 自定义模型和自定义代理的实际应用：行业用例和端到端演示.pdf

1、Orlando,FLOctober 69IBM TechXchange 2025Suhas KashyapIBM,Product Management watsonx2769:Custom models and custom agents in action:Industry Use Cases and End-to-End DemosWhat you will learn010203Model customization essentialsFine-tuning with LoRA in watsonx.aiModel customization Vs Enhanced RAG techn

2、iques When to choose whatIBM TechXchange|2025 IBM CorporationOff-the-shelf models are not exposed to your private data and unique operational workflowsNeeds to develop specialized GenAI and Agentic AI solutions that needs nuanced judgement hitting accuracy,latency and cost targetsDeploy smarter,cust

3、om AI with confidenceMeet Danny The Enterprise GenAI DeveloperDanny wants to build cost-efficient,high-performance model stacks tailored to specific uses at enterprise scaleDanny,Enterprise GenAI DevChallengeChallengeNeedsNeedsGoalsGoalsHow Danny customizes models with watsonx.ai4Wants to add enterp

4、rise data into models to achieve domain specific accuracy,low latency and low cost per querySelects best fit model for defined objectiveSelectsIngests ExtractsChunksFine Tuning ORRAG based on defined objectiveUser selected metric to best meet defined objectiveDanny,Enterprise GenAI Dev1.Defines 1.De

5、fines objectiveobjective2.Chooses a 2.Chooses a base modelbase model3.Prepares 3.Prepares datadata4.Customize 4.Customize modelmodel5.Evaluate 5.Evaluate effectivenesseffectivenessAPI enabled,scalable deployment cloud or on-prem6.Deploy 6.Deploy modelmodel7.Monitor and 7.Monitor and improveimproveUs

6、age analytics,feedback loop,rebuilding as neededPrompt LabModel GatewayText extraction APIData simulationTuning StudioCustom foundation modelAutoAI for RAGAI ServiceEvaluation StudioFully AutomatedCustom BuildCLI Templates5Context CustomizationModel CustomizationData complexity/costPrompt Engineerin