人工智能论文英文版-SMALL MODELS, BIG SUPPORT：A LOCAL LLM FRAMEWORK FOR TEACHER-CENTRIC CONTENT CREATION AND ASSESSMENT USING RAG AND CAG.pdf

SMALLMODELS,BIGSUPPORT:ALOCALLLMFRAMEWORK

FORTEACHER-CENTRICCONTENTCREATIONANDASSESSMENT

USINGRAGANDCAG

ZarreenReza∗AlexanderMazurMichaelT.Dugdale

JACOBBJACOBBJohnAbbottCollege

5Montreal,CanadaMontreal,CanadaMontreal,Canada

2zarreen.reza@jacobb.aialexander.mazur@jacobb.aimichael.dugdale@johnabbott.qc.ca

0

2

RobinRay-Chaudhuri

nJohnAbbottCollege

uMontreal,Canada

J

robin.ray-chaudhuri@johnabbott.qc.ca

6

]ABSTRACT

Y

CWhileLargeLanguageModels(LLMs)areincreasinglyutilizedasstudent-facingeducationalaids,

s.theirpotentialtodirectlysupporteducators,particularlythroughlocallydeployableandcustomizable

copen-sourcesolutions,remainssignificantlyunderexplored.Manyexistingeducationalsolutionsrely

[oncloud-basedinfrastructureorproprietarytools,whicharecostlyandmayraiseprivacyconcerns.

Regulatedindustrieswithlimitedbudgetsrequireaffordable,self-hostedsolutions.Weintroducean

1end-to-end,open-sourceframeworkleveragingsmall(3B-7Bparameters),locallydeployedLLMs

v

5forcustomizedteachingmaterialgenerationandassessment.Oursystemuniquelyincorporates

2aninteractiveloopcrucialforeffectivesmall-modelrefinement,an