Ƙamus

Retrieval-Augmented Generation (RAG)

Retrieval-augmented generation wata dabara ce da ke grounding output na language model cikin source documents da aka retrieve maimakon dogaro da parametric memory kadai. System yana fetch relevant passages daga knowledge base, yana ba su matsayin context, sannan ya umarci model ya amsa da wannan evidence kadai.

Kalmomi masu kama: RAG, retrieval augmented generation, grounded generation, context augmentation

Retrieval-augmented generation yana raba abin da model ya sani da abin da aka horar da shi da shi. A lokacin query, retriever yana jawo chunks mafi relevant daga maintained knowledge base, sannan model yana hada amsa da aka takaita ga wannan evidence. Wannan yana rage hallucination, yana sa answers su zama auditable ta citations, kuma yana ba teams damar sabunta knowledge ba tare da retraining ba. A governed setting, RAG shi ne mechanism da ke juya tambaya zuwa grounded, cited response da mutum ko policy zai iya verify kafin a amince da ita.

Me ya sa a yi amfani da RAG maimakon fine-tuning?

RAG yana ajiye knowledge a external store da za ka iya sabunta nan take, don answers su kasance current kuma kowace claim ta trace zuwa source. Fine-tuning yana bake knowledge cikin weights, wanda ya fi jinkirin refresh kuma ya fi wuya a attribute.

Me RAG pipeline ke kunsa?

Yawanci ingestion da chunking, embedding, index don vector ko hybrid search, retriever, da generation step da ke condition model a kan retrieved passages kuma ya dawo da cited evidence.

Kalmomin da suka shafi juna

Fara kyauta Yi magana da tallace-tallace

Bincika dukkan kalmomin ƙamus

Tambayoyin da ake yawan yi

Kalmomin da suka shafi juna