LLM 评估：审核微调的 LLM 以保证输出质量.pdf

1、2024 Databricks Inc.All rights reservedLLM Evaluation:LLM Evaluation:Auditing FineAuditing Fine-Tuned Tuned LLMs for Guaranteed LLMs for Guaranteed Output QualityOutput QualityLoic Pauletto,Ph.D.Loic Pauletto,Ph.D.Data Scientist,MiraklData Scientist,Mirakl1Pierre LourdeletPierre LourdeletData Scient

2、ist,MiraklData Scientist,Mirakl2024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights reservedEmpowering clients since 2012World leading marketplace platform+450 B2B and B2C clients+200k shops8.6B$GMV in 2023Develop and industrialize AI featuresFacilitation and improvement of Client O

3、perationsPlatform Security and Fraud PreventionRevenue Growth Through Personalization SolutionsThe Data Science Team:MiradogeThe Data Science Team:Miradoge2Who are we?Who are we?22024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights reservedA mA marketplacearketplace:a platform where

4、 multiple:a platform where multiple providers sell products and servicesproviders sell products and servicesSome MiraklSome Mirakl-powered marketplacespowered marketplaces3What is a marketplace?What is a marketplace?3MarketplaceSeller 1Seller 2Seller 3Seller 4Marketplace tenantCustomerssellsmanagess

5、ellbuy2024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights reserved1.Catalog Data2.The onboarding struggle3.How to auto-structure information?4.Challenges5.Ensure Quality6.Engineering integration7.Key takeaways4Main TopicsMain Topics2024 Databricks Inc.All rights reserved5Context:Co

6、ntext:CatalogCatalog-based environment based environment 2024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights reserved6What is a catalog?What is a catalog?In a marketplace contextIn a marketplace contextTitleTitle:iPhone 15 6,1”5G 128GB BlackDescriptionDescription:This is an iPhone