RecSys、主题建模和代理：弥合 GenAI 与传统 ML 之间的鸿沟.pdf

《RecSys、主题建模和代理：弥合 GenAI 与传统 ML 之间的鸿沟.pdf》由会员分享，可在线阅读，更多相关《RecSys、主题建模和代理：弥合 GenAI 与传统 ML 之间的鸿沟.pdf（48页珍藏版）》请在三个皮匠报告上搜索。

1、2024 Databricks Inc.All rights reserved1 1RecSys,Topic Modeling,and Agents:Bridging the GenAI-Traditional ML Divide Dan PechiSenior GenAI Product Specialist Databricks2024 Databricks Inc.All rights reservedAgendaBridging the GenAI-Traditional ML DivideReinventing the Wheel?Topic ModelingRecSysAgents

2、Demo2024 Databricks Inc.All rights reservedReinventing the Wheel?2024 Databricks Inc.All rights reservedThe AI/ML Wheel Then and Now Before GenAIMLOpsModels with weights/parametersMLFlow for evaluationAccuracy,Precision,Recall,F1,etcMLLib,sklearn,PyTorch for model-buildingDeterministic outputs(mostl

3、y)Pre-training vs fine-tuningAfter GenAIPreferred Qualifications2024 Databricks Inc.All rights reservedThe AI/ML Wheel Then and Now Before GenAIMLOpsModels with weights/parametersMLFlow for evaluationAccuracy,Precision,Recall,F1,etcMLLib,sklearn,PyTorch for model-buildingDeterministic outputs(mostly

4、)Pre-training vs fine-tuningAfter GenAILLMOpsCompound systems of APIsMLFlow for evaluationLLM-as-a-judge derived metricsLangchain,MCP,DSPy,and Mosaic AINon-deterministic API outputsPre-training vs fine-tuningPreferred Qualifications2024 Databricks Inc.All rights reservedThe AI/ML Wheel Then and Now

5、Before GenAIMLOpsModels with weights/parametersMLFlow for evaluationAccuracy,Precision,Recall,F1,etcMLLib,sklearn,PyTorch for model-buildingDeterministic outputs(mostly)Pre-training vs fine-tuningAfter GenAILLMOpsCompound systems of APIsMLFlow for evaluationLLM-as-a-judge derived metricsLangchain,MC

6、P,DSPy,and Mosaic AINon-deterministic API outputsPre-training vs fine-tuning vs PEFT vs prompt-tuning vs combinations of the aboveVector DBs?Late interaction models?Agents?Tools?Prompt caching?Prompt injection?Preferred Qualifications2024 Databricks Inc.All rights reservedThe AI/ML Wheel Then and No