DecoderLens: Layerwise Interpretation of Encoder-Decoder Transformers

TL;DR

DecoderLens is a novel interpretability method for encoder-decoder Transformers that visualizes intermediate representations by enabling cross-attention to encoder layers, revealing how information flows and is processed at different depths.

Contribution

It introduces DecoderLens, a new technique for layerwise interpretation of encoder-decoder models by mapping internal states to human-understandable outputs, enhancing understanding of model internals.

Findings

Reveals specific subtasks solved at various layers

Shows information flow within encoder components

Applied successfully to multiple NLP tasks

Abstract

In recent years, many interpretability methods have been proposed to help interpret the internal states of Transformer-models, at different levels of precision and complexity. Here, to analyze encoder-decoder Transformers, we propose a simple, new method: DecoderLens. Inspired by the LogitLens (for decoder-only Transformers), this method involves allowing the decoder to cross-attend representations of intermediate encoder layers instead of using the final encoder output, as is normally done in encoder-decoder models. The method thus maps previously uninterpretable vector representations to human-interpretable sequences of words or symbols. We report results from the DecoderLens applied to models trained on question answering, logical reasoning, speech recognition and machine translation. The DecoderLens reveals several specific subtasks that are solved at low or intermediate layers, shedding new light on the information flow inside the encoder component of this important class of models.