September 2025 Summaries

Small language models (SLMs) offer a cost-effective solution to the challenges posed by the large size and high hardware demands of traditional language models, particularly when combined with CPUs for edge deployment. The blog illustrates this by detailing the optimization of Arcee's AFM-4.5B model on Intel Xeon 6 processors using the Intel OpenVINO toolkit and Hugging Face's Optimum Intel library. Released in June 2025, the AFM-4.5B model has proven effective in edge and compute-constrained environments, outperforming other models of similar size. The Intel Xeon 6, equipped with Performance and Efficiency Cores, enhances AI workload efficiency, while OpenVINO and Optimum Intel simplify model optimization and deployment. The blog provides a demonstration of deploying AFM-4.5B on an AWS EC2 instance using Docker and OpenVINO Model Server, showcasing its potential for scalable, high-performance AI solutions beyond large data centers.