Fiddler: CPU-GPU Orchestration for Fast Inference of Mixture-of-Experts Models

Large Language Models (LLMs) based on Mixture-of-Experts (MoE) architecture are showing promising performance on various tasks. However, running them on resource-constrained settings, where GPU memory resources are not abundant, is challenging due to huge model sizes. Existing systems that offload model weights to CPU memory suffer from the significant overhead of frequently moving data between CPU and GPU.In this paper, we propose Fiddler, a resource-efficient inference engine with CPU-GPU orchestration for MoE models.The key idea of Fiddler is to use the computation ability of the CPU to minimize the data movement between the CPU and GPU.Our evaluation shows that Fiddler can run the uncompressed Mixtral-8x7B model, which exceeds 90GB in parameters, to generate over $3$ tokens per second on a single GPU with 24GB memory, showing an order of magnitude improvement over existing methods.We are going to release the code of Fiddler as open-source software.