从LLM角度学习和了解MoE架构

MoE是目前国内LLM采用的技术路线，Deepseek R1、Qwen3、GLM4都采用了MoE架构。

这里尝试理解什么是MoE，以及MoE有哪些优点和缺点，后续分析为什么这么多LLM采用MoE。

1 什么是MoE

MoE，全称为Mixed Expert Models，1991年论文Adaptive Mixture of Local Exports提出。

1.1 MoE示例

在相同计算资源和参数量条件下，MoE模型与Dense模型相比，训练速度更快。

也就是说，MoE由于训练效率的提升，可以是更大的模型。

比如Google Switch Transformer模型，大小是T5-XXL的15倍，相比T5-XXL模型快4倍达到固定困惑度PPL。

In deep learning, models typically reuse the same parameters for all inputs. Mixture of Experts (MoE) models defy this and instead select different parameters for each incoming example. The result is a sparsely-activated model---with an outrageous number of parameters---but a constant computational cost. However, despite several notable successes of MoE, widespread adoption has been hindered by complexity, communication costs, and training instability. We address these with the introduction of the Switch Transformer. We simplify the MoE routing algorithm and design intuitive improved models with reduced communication and computational costs. Our proposed training techniques mitigate the instabilities, and we show large sparse models may be trained, for the first time, with lower precision (bfloat16) formats. We design models based off T5-Base and T5 Large (Raffel et al., 2019) to obtain up to 7x increases in pre-training speed with the same computational resources. These improvements extend into multilingual settings where we measure gains over the mT5-Base version across all 101 languages. Finally, we advance the current scale of language models by pre-training up to trillion parameter models on the "Colossal Clean Crawled Corpus", and achieve a 4x speedup over the T5-XXL model.

https://arxiv.org/pdf/2101.03961

在相同计算预算条件下，MoE可以显著扩大模型的规模。特别是在预训练阶段，与稠密模型相比，混合专家模型通常能够更快地达到相同的质量水平。