Enhancing Instruction-Following Capabilities in Seq2Seq Models: DoLA Adaptations for T5
This work addresses instruction-following in seq2seq models for NLP applications, but it is incremental as it applies an existing method to a new architecture.
The paper adapted the DoLa contrastive decoding method to T5 and FLAN-T5 encoder-decoder models to enhance instruction-following capabilities, finding that it improves faithfulness in some task categories but harms others, with a layer-by-layer analysis to quantify its impact on token probabilities.
Contrastive decoding is a lightweight and effective inference-time method that improves the quality of text generation in Large Language Models. However, algorithms such as DoLa (Decoding by Contrastive Layers) have only been implemented in decoder-only architectures and studied for their impact on improving factuality. This work adapts DoLa for the T5 and FLAN-T5 model families and evaluates its impact on the models' instruction following capabilities, which to our knowledge is the first implementation of a contrastive decoding strategy in an encoder-decoder architecture. Our results show that DoLa improves the faithfulness of text generation for certain categories of tasks and harms others. To understand these results, we present a layer-by-layer analysis of logit evolution in a FLAN-T5 model to quantify DoLa's impact on token output probabilities.