Closed Fansomone closed 3 years ago
Cppflow (libtensorflow to be exact) automatically uses your GPU if it is available. Please check the error logs to see if you have incompatible driver/cudatookit.
You can try export CUDA_VISIBLE_DEVICES=0
, but it is often not necessary.
thanks a lot,my problem has been solved
My model (h5 format) can be predicted using GPU, but when I convert it to pb format, I can't call GPU to predict using CppFlow. What should I do?
I set up the following configuration in the model.py file
os.environ["CUDA_VISIBLE_DEVICES"] = "0"