Optimizing the Differentiable Search Index (DSI) with data augmentation (Num2Word, Stopwords Removal, POS-MLM) and parameter-efficient fine-tuning (LoRA, QLoRA, AdaLoRA, ConvoLoRA), improving retrieval accuracy and efficiency while reducing memory and computational overhead. Evaluated on the MS MARCO dataset for scalable performance.
-
Updated
Apr 3, 2025 - Jupyter Notebook