search

icenet-ai / icenet-pipeline

The icenet-pipeline repository illustrates operational execution of the IceNet model
MIT License
3 stars 5 forks source link

Pipeline tensorflow installation, conda or other way. #32

Closed matscorse closed 8 months ago

matscorse commented 9 months ago

There is a note in icenet installation:

Bear in mind when installing icenet (and by dependency tensorflow) you will need to be on a CUDA/GPU enabled machine for binary linkage. As per current (end of 2022) tensorflow guidance do not install it via conda.

So when performing " pip install -e . " I assume this isn't using conda but is using pip. If I run this on a non gpu workstation is my install broken?

JimCircadian commented 9 months ago

Hi @matscorse: if you run pip within a conda environment, that's fine. It means that you'll be leveraging the python install within that conda environment. In the conda environment created by icenet-pipeline, and on the BAS HPC specifically, the GPU libraries are all available so you don't need to worry too much about linking to cuda etc, it'll happen based on the requirements specified in the python project.

If you continue through the instructions you should see that the whole thing works with a GPU when you get to the training step. If it doesn't we can reinstall tensorflow on a GPU enabled machine and make sure, but that probably won't be the case. Either way, assume your environment is fine until something suggest otherwise! :)

EDIT pip and conda aren't mutually exclusive 😉

Hope that helps!

bnubald commented 8 months ago

Closing as solved, please do feel free reopen if you needed more clarification @matscorse!