GeNetE - Generative Network Environments: A Specialized Small Language Model

Network emulation is crucial for testing and validating network systems, especially in the military domain involving devices from multiple network vendors, interfaces, and components in a federated mission. However, constructing the network infrastructure generally requires specialized expertise and a time-consuming setup process. Although Large Language Models (LLMs) can assist in code generation, general-purpose models often lack the domain-specific precision required for network emulators. This article introduces GeNetE, a Generative Network Environments framework, for creating a Specialized Small Language Model (SSLM) by fine-tuning open-weight LLMs using synthetic, domain-specific datasets. GeNetE enables streamlined emulator use, runs efficiently on low-resource devices, and reduces user workload. GeNetE was extensively evaluated using the Mininet emulator, employing various model sizes and quantization techniques, achieving approximately 60% valid Software-Defined Network environment generation with a compact 2.4 GB model, underscoring its potential for accessible and efficient network testing.