abhmul / InstallCUDA-Kerasv2

A Fuss-Free 3-step Guide to install updated Tensorflow V1 and Keras V2 on Ubuntu 16.04 with a GPU

MIT License

8 stars 4 forks source link

Changed tensorflow installation to v1.2.0 instead of latest v1.3.0 to… #1

Closed gunnyliu closed 7 years ago

gunnyliu commented 7 years ago

… maintain compatibility with CuDNN 5.1

Basically, recent upgrade to TF v1.3.0 broke our CuDNN

For patching, install v1.2.0 instead using pypi wheel files
To review TF v1.3.0 and change CuDNN to 6.0 soon

abhmul commented 7 years ago

I'll merge this for now, and work on updating this so we can still use the latest version of Tensorflow. Basically, we'll just have to fix this every time Tensorflow updates for the new Cudnn.