Dynamic Activation Pitfalls in LLaMA Models: An Empirical Study
This work identifies pitfalls in dynamic activation for LLaMA models, which is incremental as it analyzes existing methods rather than introducing new ones.
The study investigated dynamic activation mechanisms in LLaMA models and found that they often underperform compared to ReLU counterparts, especially at high sparsity ratios, due to issues like complex prediction and inadequate sparsity.
In this work, we systematically investigate the efficacy of dynamic activation mechanisms within the LLaMA family of language models. Despite the potential of dynamic activation methods to reduce computation and increase speed in models using the ReLU activation function, our empirical findings have uncovered several inherent pitfalls in the current dynamic activation schemes. Through extensive experiments across various dynamic activation strategies, we demonstrate that LLaMA models usually underperform when compared to their ReLU counterparts, particularly in scenarios demanding high sparsity ratio. We attribute these deficiencies to a combination of factors: 1) the inherent complexity of dynamically predicting activation heads and neurons; 2) the inadequate sparsity resulting from activation functions; 3) the insufficient preservation of information resulting from KV cache skipping. Our analysis not only sheds light on the limitations of dynamic activation in the context of large-scale LLaMA models but also proposes roadmaps for enhancing the design of future sparsity schemes.