Neural ODE Transformers: Analyzing Internal Dynamics and Adaptive Fine-tuning
This provides a novel interpretability method for transformer architectures, though it appears incremental in scope.
The paper tackles the problem of understanding transformer inner workings by modeling them with neural ODEs, finding that eigenvalue analysis challenges weight-sharing assumptions and achieving comparable or better performance than vanilla transformers.
Recent advancements in large language models (LLMs) based on transformer architectures have sparked significant interest in understanding their inner workings. In this paper, we introduce a novel approach to modeling transformer architectures using highly flexible non-autonomous neural ordinary differential equations (ODEs). Our proposed model parameterizes all weights of attention and feed-forward blocks through neural networks, expressing these weights as functions of a continuous layer index. Through spectral analysis of the model's dynamics, we uncover an increase in eigenvalue magnitude that challenges the weight-sharing assumption prevalent in existing theoretical studies. We also leverage the Lyapunov exponent to examine token-level sensitivity, enhancing model interpretability. Our neural ODE transformer demonstrates performance comparable to or better than vanilla transformers across various configurations and datasets, while offering flexible fine-tuning capabilities that can adapt to different architectural constraints.