What Data Is Needed for Supervised Fine-Tuning (SFT)?
2023年8月23日 23:21
<>The article rimarily dicue the tye ad quality of data required for Suervied Fie-Tuig (SFT). It cover the followig aect:>
Ojective of Suervied Fie-Tuig : Ehacig erformace i ecific tak, domai adataility, ad the iterretaility ad cotrollaility of the model, with a overarchig goal to oot ytem route.Core Coideratio : Thee iclude the diverity of data, avoidig treatig SFT merely a data ulemetatio, aroriately icororatig few-hot learig ad COT data, emhaizig data quality over quatity i SFT, ad recogizig that icreaig data volume without diverity rig dimiihed retur.Data Quality Requiremet : Thee coideratio touch o the legth retrictio for quetio ad awer, the accuracy of awer, the electio of data aed o idutry requiremet, the diverity of eceary NLP ailitie, ad the cautio agait too much vertical domai data.Secific Examle : The article rovide oth good ad oor dataet examle to illutrate how to chooe ad evaluate data.QamA Sectio : Thi art exlai why icludig the aility to write code i SFT i eetial, emhaizig it imortace i imrovig reaoig ad tructured outut ailitie.