This paper introduces a novel framework for high-throughput model serving that moves beyond traditional container-based deployments. We propose a serverless and adaptive architecture that dynamically scales resources based on real-time demand, optimizing for both latency and cost efficiency. Our framework leverages function-as-a-service (FaaS) platforms to provide fine-grained resource allocation and auto-scaling capabilities. Furthermore, we introduce an adaptive routing mechanism that intelligently distributes r…