Relion 5.0.0 container does not work with Blush, etc

Accelerated Computing NGC GPU Cloud Container: HPC
1

HI,

we are using the relion 5.0.0 container and relion fails when using the --blush option.
We checked and we could not find the required packages inside the container.

The error:

level=info msg="Setting CUDA_MODULE_LOADING=LAZY. Please see https://docs.nvidia.com for more information."

Something went wrong in the external Python call...
Command: relion_python_blush Class3D/job097/run_it001_class001_external_reconstruct.star  --gpu , 

---------------------------------- PYTHON ERROR ---------------------------------
   Has RELION been provided a Python interpreter with the correct environment?
   The interpreter can be passed to RELION either during Cmake configuration by
     using the Cmake flag -DPYTHON_EXE_PATH=<path/to/python/interpreter>.
   NOTE: For some modules TORCH_HOME needs to be set to find pretrained models
---------------------------------------------------------------------------------

  Using python executable:

The command I ran:

singularity  run --env PATH=/usr/local/relion/sm90/bin/:$PATH --nv $SIF_FILE /usr/local/openmpi/bin/mpirun -np 5  relion_refine_mpi --o Class3D/job097/run --i Select/job014/particles.s
tar --ref InitialModel/job015/initial_model.mrc --trust_ref_size --ini_high 50 --dont_combine_weights_via_disc --pool 30 --pad 1  --ctf --iter 25 --tau2_fudge 4 --particle_diameter 200
 --K 4 --flatten_solvent --zero_mask --blush  --oversampling 1 --healpix_order 2 --offset_range 5 --offset_step 2 --sym C1 --norm --scale  --j 4 --gpu ""  --pipeline_control Class3D/jo
b097/

Any Idea how to resolve this ?

2

Hi Greta,
As blush depends on python support a new container image will be required. We’ll consider adding that into the next NGC release; In the meantime you may consider modifying the Dockerfile located within the container,/usr/src/Dockerfile, to create your own custom Relion container image.

Best,
Adam Simpson
Performance Lab | NVIDIA

3

Hi Adam,

thanks for your fast reply.
Yeah I/we have been trying to build relion/5.0.0 on Grace Hoppers with and without containers for some time and the python requirements by relion fail on the aarch64 architecture. That is why I was hoping that Nvidia could provide a recipe for that…

Thanks anyway,
Greta