If I use 'pip install quiver_engine' diretly, it show check precodition and install tensorflow-1.4.0(not gpu version) again while I had already installed a good tensorflow-gpu-1.4.0. And keras will use default of TF and not use TF-GPU, so I have to uninstall tensorflow-1.4.0 first, then I find tensorflow-gpu-1.4.0 is BAD. I have to install it again(after quiver_engine).
So I submit this issue to indicate that use 'pip install quiver_engine' will cause python system check and install tf(CPU version) no matter if there is a GOOD TF(GPU version), it is a little bit inconvenince.
If possible, pls check and solve.
but at least I can install quiver_engine first and REINSTALL tf-gpu AGAIN, it works.
thks
If I use 'pip install quiver_engine' diretly, it show check precodition and install tensorflow-1.4.0(not gpu version) again while I had already installed a good tensorflow-gpu-1.4.0. And keras will use default of TF and not use TF-GPU, so I have to uninstall tensorflow-1.4.0 first, then I find tensorflow-gpu-1.4.0 is BAD. I have to install it again(after quiver_engine). So I submit this issue to indicate that use 'pip install quiver_engine' will cause python system check and install tf(CPU version) no matter if there is a GOOD TF(GPU version), it is a little bit inconvenince. If possible, pls check and solve. but at least I can install quiver_engine first and REINSTALL tf-gpu AGAIN, it works. thks