Update Tensorboard while keeping Tensorflow old with conda

Question

I have some legacy Keras/Tensorflow code, which is unstable using latest Tensorflow versions (1.13+). It works just fine with previous versions. However i want to use Pytorch's Tensorboard support which requires it to be 1.14+. I've installed all Tensorflow-related packages to 1.10 and wanted to do just conda install tensorboard=1.14 but it removes tensorflow=1.10 as a requirement. I know that these packages are generally independent. How to upgrade tensorboard while keeping tensorflow old? Preferably i would like to use a single conda environment.

Answer 1

正如@jdehesa在评论中建议的那样，最好为pytorch使用不同的conda环境，然后在其中安装tb
!pip install tb-nightly