Created by: guomingz

We developed the calibration tool for int8 inference. Here is the document for this tool.

Introduction of Accuracy Calibration Tool for 8 Bit Inference

The 8-bits (INT8) inference is also known as the Low Precision Inference which could speed up the inference with the lower accuracy loss. It has higher throughput and lower memory requirements compared to FP32. As the PaddlePaddle enables the INT8 inference supporting, we release a accuracy tool(Calibration.py) at the same time. This tool will generate the quantization parameters and quantized model file finally.