On the latest amazon deep learning ami (v3) I had to use
echo -e "[global]\ndevice = cuda\nfloatX = float32\n[gpuarray]\npreallocate = 1" > /home/ec2-user/.theanorc
,
install libpygpu and restart the kernel to get it working
conda install pygpu
Restart kernel
in which directory do I intall? I tried conda install pygpu in home directory and got error No writable package cache directories found in
(uā/home/daksh/anaconda2/pkgsā, uā/home/daksh/.conda/pkgsā)
Could you help?
It think itās a permission issue. Try entering sudo conda install pygpu
im having the same problem. working with a paperspace unbuntu 16 box.
process:
git clone repo
run bash install-gpu.sh from setup folder
ssh into server and open jupyter note book
initial error is:
so I did conda install pygpu
and then https://stackoverflow.com/questions/44765376/valueerror-you-are-trying-to-use-the-old-gpu-back-end-when-importing-keras
aka ran export THEANO_FLAGS=mode=FAST_RUN,device=cuda,floatX=float32
then the above attribute error happens.
if I restart kerenel im back to the error in the image againā¦
very frustrating. Iāve spent half the day trying to fix thisā¦ plz plz help!
In the Setup video, near the very end (around 1:28:00) Jeremy shows a couple of resource files. ~/.theanorc has a ādevice=ā setting. I changed this to ādevice = cuda*ā and that took care of the error.
Iām now getting the āAttributeError: (āThis name is already takenā, āfloatXā)ā and canāt figure out how to fix.
Ideas?
Hey folks,
In case you are not aware. you should probably switch to version 2 of the course (updated and now uses pytorch).
Itās a much easier setup process and works by just running the setup script.
Btw. Theano has decided to stop active development : (
I expect the course to be heavy on pytorch and tensorflow in the future.