Closed jguhlin closed 1 year ago
I'm confused, if you can use singularity presumably you can build your own native singularity image using either the PyPI or conda packages?
Not sure what there is to be confused about. Would much prefer to use the official container if that is an option, as it is a workshop.
Doesn't look like the conda version supports the GPU either:
https://github.com/nanoporetech/medaka/issues/165#issuecomment-634003709
We've been able to resolve the issue with a built-from-source module.
There is an error in the README, Medaka does not use TensorFlow 2.2, it uses 2.7. Which has different requirements for CUDA & related libs.
https://github.com/nanoporetech/medaka/blob/master/requirements.txt#L8
It would still be awesome to have a working docker to singularity pipeline working to save time (Conda is still solving the environment, going on 20 minutes now, even though it looks like it won't work with the GPU anyways, so won't work for us). I'll probably futz around with it a bit more, one org can use singularity but not conda, and will report if I get anywhere. Feel free to close this issue.
Built container using:
As docker is not allowed on most research instances. mini_align step works fine, but running the consensus step I get the following error.
The model definitely exists:
Installing via conda isn't an option (workshop), so a container seems to be the best option. Any ideas?