MoE Jetpack：从稠密检查点到视觉任务自适应混合专家模型
MoE Jetpack: From Dense Checkpoints to Adaptive Mixture of Experts for Vision Tasks
AI总结：
MoE Jetpack 提出通过检查点回收和 SpheroMoE 层将现有稠密检查点高效微调为混合专家模型，在视觉任务上显著提升收敛速度与准确率。
摘要：
稀疏激活的混合专家（MoE）模型为传统的稠密激活（dense）模型提供了一种有前景的替代方案，能够同时提升质量和计算效率。然而，从零开始训练 MoE 模型需要大量的数据和计算资源。此外，像 timm 这样的公共仓库主要提供预训练的稠密检查点，缺乏针对 MoE 模型的类似资源，阻碍了其应用推广。为了弥合这一差距，我们提出了 MoE Jetpack，一种将稠密检查点微调为 MoE 模型的有效方法。MoE Jetpack 包含两项关键技术：（1）检查点回收（checkpoint recycling），将稠密检查点重新用作 MoE 模型的初始权重，从而加速收敛、提高准确率，并减轻预训练的计算负担；（2）超球面自适应 MoE（SpheroMoE）层，优化 MoE 架构以更好地整合稠密检查点，提升微调性能。我们在视觉任务上的实验表明，MoE Jetpack 在将稠密检查点微调为 MoE 模型时，显著提高了收敛速度和准确率。我们的代码将在 https://github.com/Adlith/MoE-Jetpack 公开发布。
英文摘要：
The sparsely activated mixture of experts (MoE) model presents a promising alternative to traditional densely activated (dense) models, enhancing both quality and computational efficiency. However, training MoE models from scratch demands extensive data and computational resources. Moreover, public repositories like timm mainly provide pre-trained dense checkpoints, lacking similar resources for MoE models, hindering their adoption. To bridge this gap, we introduce MoE Jetpack, an effective method for fine-tuning dense checkpoints into MoE models. MoE Jetpack incorporates two key techniques: (1) checkpoint recycling, which repurposes dense checkpoints as initial weights for MoE models, thereby accelerating convergence, enhancing accuracy, and alleviating the computational burden of pre-training; (2) hyperspherical adaptive MoE (SpheroMoE) layer, which optimizes the MoE architecture for better integration of dense checkpoints, enhancing fine-tuning performance. Our experiments on vision tasks demonstrate that MoE Jetpack significantly improves convergence speed and accuracy when fine-tuning dense checkpoints into MoE models. Our code will be publicly available at https://github.com/Adlith/MoE-Jetpack.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
