ML基础架构的PYTORCH方法.docxVIP

RUNHOUSE:APYTORCHMLINFRAAPPROACHTORohinBhasinCarolineChenJune11,20241?2024

RUNHOUSE:APYTORCHMLINFRA

APPROACH

TO

RohinBhasinCarolineChenJune11,2024

1

?2024DatabricksInc.—Allrightsreserved

?2024DatabricksInc.—Allrights

?2024DatabricksInc.—Allrightsreserved

MODERN

ML

INFRA

IS

FRAGMENTED

Workflowsareinflexible,duplicative,andunreproducible.

2

?2024DatabricksInc.—Allrightsreserved

PRODUCTIONDAGs,orchestrators,containersTRANSLATIONPACKAGING1-4monthprocesstranslatingcodeforspecificinfra,learningDSLspackagingcodeintoDAGscontainerization