RVTensor: A Light-Weight Neural Network Inference Framework Based on the RISC-V Architecture.

The open-source instruction set architecture RISC-V has developed rapidly in recent years, and its combination mode of multiple sub-instruction sets has attracted the attention of IoT vendors. However, research on the IoT scenario inference framework based on the RISC-V architecture is rare. Popular frame-works such as MXNet, TensorFlow, and Caffe are based on the X86 and ARM architectures, and they are not optimized for the IoT scenarios. We propose RVTensor that a light-weight neural network inference framework based on the RISC-V architecture. RVTensor is based on the SERVE.r platform and is optimized for resource-poor scenarios. Our experiments demonstrate that the accuracy of RVTensor and the Keras is the same.