Adaptive Head Budgeting for Efficient Multi-Head Attention
For practitioners of Transformer models, this provides a principled method to improve efficiency without sacrificing performance, though it is incremental over existing adaptive attention mechanisms.
BudgetFormer introduces adaptive multi-head attention that dynamically selects the number and relevance of heads per input, reducing FLOPs and memory while matching or surpassing standard full attention on text classification tasks.
Transformers have become the dominant architecture across a wide range of domains, largely due to the effectiveness of multi-head attention in capturing diverse representation subspaces. However, standard multi-head attention activates all heads uniformly for every input, regardless of task requirements or input complexity. In many scenarios, particularly for coarse-grained tasks such as text classification, the relevant information is often global and does not require the full diversity of attention heads. As a consequence, using a fixed number of heads can introduce unnecessary computational cost or lead to suboptimal performance when the allocation does not match the input. To address this limitation, we introduce BudgetFormer, a Transformer architecture equipped with an adaptive multi-head attention mechanism that dynamically allocates computational resources. Our approach learns, for each input, both a head budget corresponding to the number of attention heads required, and a relevance distribution that selects the most informative heads. We also propose a training strategy based on an exploration and exploitation trade-off, allowing the model to discover effective head configurations before converging to efficient usage patterns. Experiments on text classification tasks of varying complexity show that our method reduces inference cost in terms of FLOPs and memory, while also achieving performance that can surpass standard full multi-head attention. These results highlight the potential of adaptive head allocation as a principled approach to improving both efficiency and effectiveness in Transformer models.