Who Says Elephants Can't Run: Bringing Large Scale MoE Models into Cloud Scale Production
This work addresses the problem of inefficient inference and high memory requirements for deploying large MoE models in cloud production, enabling a paradigm shift from distilling models to direct deployment.
The paper tackles the challenge of deploying large-scale Mixture of Experts (MoE) models in production by introducing an efficient inference framework that reduces memory usage and speeds up computation, achieving up to 26x throughput speed-up, reducing model size to one-eighth via 4-bit quantization, and enabling deployment of 136x larger models with 27% less cost and better quality.
Mixture of Experts (MoE) models with conditional execution of sparsely activated layers have enabled training models with a much larger number of parameters. As a result, these models have achieved significantly better quality on various natural language processing tasks including machine translation. However, it remains challenging to deploy such models in real-life scenarios due to the large memory requirements and inefficient inference. In this work, we introduce a highly efficient inference framework with several optimization approaches to accelerate the computation of sparse models and cut down the memory consumption significantly. While we achieve up to 26x speed-up in terms of throughput, we also reduce the model size almost to one eighth of the original 32-bit float model by quantizing expert weights into 4-bit integers. As a result, we are able to deploy 136x larger models with 27% less cost and significantly better quality compared to the existing solutions. This enables a paradigm shift in deploying large scale multilingual MoE transformers models replacing the traditional practice of distilling teacher models into dozens of smaller models per language or task.