从不确定性到确定性：确定性LLMOps的策略.pdf

1、2024 Databricks Inc.All rights reservedAmanda Milberg,Dataiku Amanda Milberg,Dataiku June 2024June 20241From Uncertainty From Uncertainty to Certainty:to Certainty:Strategies for Deterministic LLMOps2024 Databricks Inc.All rights reservedThe LLM LandscapeStrategies for LLMOpsProduct DemoThe LLM Land

2、scapeStrategies for LLMOpsProduct Demo2Discussion PointsDiscussion PointsDiscuss the growing LLM Landscape from 2018-presentOutline key factors when building a LLM ecosystem to meet business needsDefine the term LLMOpsDifferences between AI/ML vs LLMs workflowsCommon problems and proposed solutions

3、for monitoring LLMs Discuss an illustrative use case building a RAG application in DataikuMock up a LLMOps solution based on strategies disucssed 2024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights reserved3The The LLMLLMLandscapeLandscapeHow many Large How many Large Language Mode

4、ls Language Models have been developed have been developed and released to and released to date?date?2024 Databricks Inc.All rights reserved5The Growing LLM Landscape The Growing LLM Landscape There are over 125 LLMs available in the model landscape There are over 125 LLMs available in the model lan

5、dscape+120%+120%+40%+40%Source:https:/ from 2022 to 2022 to 2023 2023 Expectation is that model release count will continue to continue to increaseincrease Likley churnchurn of some models with newer performannt models replacing old ones Witnessing a race of race of innovation innovation between the

6、 major players Models are getting bigger,bigger,better,and multibetter,and multi-modal modal Advancements in both the open source community and open source community and LLM Providers LLM Providers 2024 Databricks Inc.All rights reservedChoose a models that is adaptive to your needsSelf Hosted vs.AP