Hi, I wrote a c++ program using libtorch to test the torchscript models, I found the speed of fp16 model is higher than fp32,it is exciting, but I only test one pair fg/bg image, I can t see any difference in the results between fp16 and fp32 models, so can you tell me how much does the quality will decrease of fp16 model?
Hi, I wrote a c++ program using libtorch to test the torchscript models, I found the speed of fp16 model is higher than fp32,it is exciting, but I only test one pair fg/bg image, I can t see any difference in the results between fp16 and fp32 models, so can you tell me how much does the quality will decrease of fp16 model?