从业者对LLMOPS的看法.pdf

1、2024 Databricks Inc.All rights reserved1A Practitioners A Practitioners perspective on perspective on LLMOpsLLMOpsHien LuuHien Luu06/12/202406/12/20242024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights reserved22024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights r

2、eservedImplementing GenAI at scale in the enterprise is like teenage sex:everyone talks about it,nobody really knows how to do it,everyone thinks everyone else is doing it,so everyone claims they are doing itCassie Kozyrkov32024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights reserv

3、ed4About MeAbout MeHead of ML Platform DoorDashHien LuuPrior work experience:2024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights reserved Unique Challenges of LLMs Application LLMs Application Archetypes LLM Use Cases Practitioners Perspective on LLMOps5AgendaAgenda2024 Databricks

4、Inc.All rights reserved2024 Databricks Inc.All rights reserved6Is MLOps Still Relevant?Is MLOps Still Relevant?DevOpsReproducibilityCollaborationMLOpsAutomationObservabilityCI/CDTools and Best Practices2024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights reserved7GenAI AdvancementGe

5、nAI AdvancementCapabilityTimeContinue to improve(speed,performance,cost)2024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights reserved8LLM Application Unique Challenges LLM Application Unique Challenges ClassicalML ModelsLLMs(FM)vs2024 Databricks Inc.All rights reserved2024 Databrick

6、s Inc.All rights reserved9LLM Application Unique Challenges LLM Application Unique Challenges LLMs(FM)General purpose tasksLess end-2-end controlCompute is the new oilGPUs are a mustHallucinationPrompt Engineering is the New Programming Language2024 Databricks Inc.All rights reserved2024 Databricks