hi, I modify the python example (/workspace/tensorrt/sample/python/int8_caffe_mnist) to serialize a int8 engine with 7000 pictures, but the GPU (V100) is never used when building the engine by the method of “build_cuda_engine()”. what’s wrong with it and how can I use the GPU to accelerate calibrate the int8 model?
Hello langlxd,
This seems to be an interesting issue. Can you send me a repro package so I can look into it and see if it’s a bug?