On Zero-Initialized Attention: Optimal Prompt and Gating Factor Estimation
University of Science, VNU-HCM, Ho Chi Minh City, Vietnam · The University of Texas at Austin · VinAI · QualComm · University of Stuttgart and Max Planck Research School for Intelligent Systems, Germany · German Research Center for AI · Universität Stuttgart · University of Texas at Austin
PDF 由论文原始站点提供，PaperCompass 不保存论文文件。
摘要
LLaMA-Adapter has recently emerged as an efficient fine-tuning technique for LLaMA models, leveraging zero-initialized attention to stabilize training and enhance performance. However, despite its empirical success, the theoretical foundations of zero-initialized attention remain largely unexplored. In this paper, we provide a rigorous theoretical analysis, establishing a connection between zero-initialized attention and mixture-of-expert models. We prove that both linear and non-linear prompts, along with gating functions, can be optimally estimated, with non-linear prompts offering greater flexibility for future applications. Empirically, we validate our findings on the open LLM benchmarks, demonstrating that non-linear prompts outperform linear ones. Notably, even with limited training data, both prompt types consistently surpass vanilla attention, highlighting the robustness and adaptability of zero-initialized attention.