Skip to content
Snippets Groups Projects
Normal view Permalink
benchmark.txt 1.13 KiB
Newer Older
  • Learn to ignore specific revisions
    • Xuan Gu's avatar
    Update file benchmark.txt
    Xuan Gu committed
    1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 
    

MODULE_NAME=nnunet_for_pytorch
MODULE_VERSION=21.11.0
WORK_DIR=/proj/nsc_testing/xuan/berzelius-benchmarks/PyTorch/Segmentation/nnUNet
CONTAINER_DIR=/proj/nsc_testing/xuan/containers/${MODULE_NAME}_${MODULE_VERSION}.sif

mkdir -p $WORK_DIR/data $WORK_DIR/results


export APPTAINER_BINDPATH=$WORK_DIR/data:/data,$WORK_DIR/results:/results


To download and preprocess the data run:

apptainer exec --nv --pwd /workspace/nnunet_pyt $CONTAINER_DIR python download.py --task 01  

apptainer exec --nv --pwd /workspace/nnunet_pyt $CONTAINER_DIR  python /workspace/nnunet_pyt/preprocess.py --task 01 --dim 2
 
Start benchmarking:

apptainer exec --nv --pwd /workspace/nnunet_pyt $CONTAINER_DIR python scripts/benchmark.py --mode train --gpus 1 --dim 2 --batch_size 256 --amp

apptainer exec --nv --pwd /workspace/nnunet_pyt $CONTAINER_DIR python scripts/benchmark.py --mode predict --gpus 1 --dim 2 --batch_size 256 --amp


# Known issue
    Xuan Gu's avatar
    Update 3 files  
    Xuan Gu committed
    28 
    # ImportError: cannot import name 'get_num_classes' from 'torchmetrics.utilities.data' (/opt/conda/lib/python3.8/site-packages/torchmetrics/utilities/data.py)
    Xuan Gu's avatar
    Update file benchmark.txt
    Xuan Gu committed
    29 
    https://github.com/NVIDIA/DeepLearningExamples/issues/1113