Latency Dataset for Once-for-All (OFA) Network on Ultra96-V2 FPGA

This dataset presents detailed latency measurements for all MobileNetV3 building blocks used in the Once-for-All (OFA) neural architecture framework, targeting deployment on edge FPGAs. The measurements were obtained using the Xilinx Vitis AI Profiler on the Ultra96-V2 development board under int8 quantization. The dataset includes latency profiles for individual OFA MobileNetV3 blocks as well as batch profiling results, generated within a PetaLinux runtime environment. This resource allows for accurate latency estimation during neural architecture search (NAS) to find efficient, deployment-ready deep learning models based on MobileNetv2 search space for edge inference on FPGA platforms.