<p>Fog Radio Access Network (Fog RAN) has recently emerged as a promising architecture for supporting low-latency applications by bringing fog nodes and cloud resources closer to end users. However, existing research on computational offloading in Fog RAN lacks a comprehensive framework that addresses three key aspects: where to offload tasks, which processing nodes to utilize, and how to allocate resources for tasks with varying latency requirements. To address this gap, we propose TOFRA (Task Offloading for Fog RAN), a novel latency-aware task offloading framework. TOFRA is a centralized system that determines the optimal offloading strategy, whether to execute tasks locally, at fog nodes, or in the cloud, selects the appropriate processing servers, and intelligently allocates both radio and computation resources based on task priorities. We employ a Genetic Algorithm (GA) to solve this optimization problem, as it effectively handles large-scale and complex system scenarios. Simulation results demonstrate that TOFRA significantly reduces latency, particularly by minimizing maximum latency for high-priority tasks compared to existing approaches, while maintaining acceptable performance for lower-priority tasks. The algorithm also exhibits fast convergence, typically stabilizing within 25 iterations, even in networks with up to 210 users. In summary, TOFRA represents a substantial advancement in task offloading strategies for Fog RAN systems across diverse network loads. Future work will extend the framework by incorporating factors such as packet loss, dynamic network failures, and multi-objective optimization, including energy consumption and bandwidth efficiency, enhancing its robustness and applicability in real-world deployments.</p>