Xilinx Vitis-AI is a platform for high-performance deep learning compression and inference on Xilinx FPGA device.
8bit per-tensor symmetric linear quantization with power of two scales.
where \(s\) is scaling factor to quantize a number from floating range to integer range, \(lb\) and \(ub\) are bounds of integer range, and [lb, ub] = [-128, 127].
Vitis-AI provides a software stack for DPUs of Xilinx FPGA, including network quantization and DPU inference etc.. To apply MQBench to Vitis-AI, we support a backend and its related observer and fake quantizer which is proven to be aligned with DPU. For now, we have supported typical operations used in classification and detection jobs: The supported Operations are listed here: