# Singularity Image kernel in Jupyter Notebook

## :memo: Where do I start?
- 聯絡窗口 Email us : [e-mail] 王小姐

## :a: Singularity 容器部屬
1. 切換到原生 Python 3 (ipykernel)
2. 更改以下三個參數
- 安裝儲存位置 saveFolder=?
- Kernel標籤名稱 label=?
- Image映像檔位置 IMAGE=?

%%bash
### configure
saveFolder=work
label=LlamaFactory
IMAGE=/work/u00cjz00/nvidia/cuda118/c00cjz00_cuda11.8_pytorch_2.1.2-cuda11.8-cudnn8-devel-llama_factory.sif

### 1. IMAGE basename
IMAGE_basename=S-${saveFolder}-${label}_$(basename "$IMAGE" .sif)

### 2. VIRTUAL LIBRARY and BINARY FOLDER
libraryFolder=/${saveFolder}/$(whoami)/libraryFolder/${IMAGE_basename}/local/lib
libraryFolder_binding=${libraryFolder}:${HOME}/.local/lib
binFolder=/${saveFolder}/$(whoami)/libraryFolder/${IMAGE_basename}/local/bin
binFolder_binding=${binFolder}:${HOME}/.local/bin
rm -rf /${saveFolder}/$(whoami)/libraryFolder/${IMAGE_basename}
mkdir -p ${libraryFolder} ${binFolder}

# 3. PIP INSTALL SLAVE IPYKERNEL
ml libs/singularity/3.10.2
singularity exec -B ${libraryFolder_binding} -B ${binFolder_binding} ${IMAGE} pip install -q ipykernel IProgress ipywidgets

# 4. IPYKERNEL for IMAGE
IPYKERNEL=/work/u00cjz00/slurm_jobs/ipykernel/t2/image_with_ipykernel_local
mkdir -p ${HOME}/.local/share/jupyter/kernels/
rm -rf ${HOME}/.local/share/jupyter/kernels/${IMAGE_basename}
cp -rf ${IPYKERNEL} ${HOME}/.local/share/jupyter/kernels/${IMAGE_basename}
chmod -R 755 ${HOME}/.local/share/jupyter/kernels/${IMAGE_basename}
IMAGE_desc=$(echo $IMAGE | sed 's_/_\\/_g')
sed -i "s/templateSIF/${IMAGE_desc}/g" ${HOME}/.local/share/jupyter/kernels/${IMAGE_basename}/kernel.json
sed -i "s/templateImage/Image_${IMAGE_basename}/g" ${HOME}/.local/share/jupyter/kernels/${IMAGE_basename}/kernel.json
sed -i "s@templateLibrayFolder@${libraryFolder_binding}@g" ${HOME}/.local/share/jupyter/kernels/${IMAGE_basename}/kernel.json
sed -i "s@templateBinFolder@${binFolder_binding}@g" ${HOME}/.local/share/jupyter/kernels/${IMAGE_basename}/kernel.json

# 5. check size
du -sh ${libraryFolder}
echo /home/$(whoami)/.local/share/jupyter/kernels/${IMAGE_basename}/kernel.json
cat /home/$(whoami)/.local/share/jupyter/kernels/${IMAGE_basename}/kernel.json

cmd="/work/opt/ohpc/Taiwania3/libs/singularity/3.10.2/bin/singularity exec --nv --cleanenv \
-B /work -B /work/u00cjz00/os/ubuntu/bin:/usr/ubuntu_bin \
-B ${libraryFolder}:/home/g00cjz00/.local/lib \
-B ${binFolder}:/home/g00cjz00/.local/bin \
${IMAGE} \
bash -c 'export PATH=\$PATH:\$HOME/.local/bin; echo \$PATH;'
"

echo ${cmd}

3. 執行以上程式碼


4. 重新restart kernel, kernel選單即會出現新的標籤


5. 切換到新kernel 執行以下指令, 確認是否支援GPU

# GPU確認
!nvidia-smi
import torch
torch.cuda.is_available()

6. 切換到新kernel 執行以下指令, 確認torch 與cuda版本

# torch and cuda version
import torch
print(torch.__version__)
print(torch.version.cuda)


## :movie_camera: 操作影片教學
{%youtube N7BwcdMYl2k %}