Bolt is a light-weight deep learning inference library developed by Huawei that aims to automate neural network deployment and achieve extreme acceleration. It supports model conversion from Caffe, ONNX, TFLite, and TensorFlow, and targets multiple platforms including ARM CPUs (v7, v8, v8.2+, v9), x86 CPUs (AVX2, AVX512), and GPUs from Mali, Qualcomm, Intel, and AMD. The library offers various inference precisions including FP32, FP16, INT8, and 1-bit quantization, and has been widely deployed across Huawei’s internal departments.