EDT5: Proposal of an Encoder-Decoder Structure Embedding Model for T5
Domestic Conferences & Theses
Authors
Joongmin Shin*, Joogyoung Jung, Junghoon Lee, Hyuk-Chul Kwon
Abstract
Analyzed the structure of T5 and proposed a model architecture that uses both the encoder and decoder for embedding, moving beyond the previous approach that only utilized the encoder for embedding and fine-tuning.
Key Contribution
Proposed encoder-decoder embedding architecture for T5, improving upon encoder-only approaches.