Jupyter in Docker: Tensorflow cannot detector GPU while Pytorch can

JupyterLab
1

Hello,
I run JupyterLab in docker with both tensorflow==2.12.0 and pytorch==2.0.1+cu118 installed. Pytorch can detector GPU, however, tensorflow failed. See bellow:

2023-05-10_17-32
2023-05-10_17-321230×1178 181 KB

The docker image was build from jupyter/scipy-notebook:lab-3.6.3:

ARG OWNER=jupyter                      
ARG BASE_CONTAINER=jupyter/scipy-notebook:lab-3.6.3
FROM $BASE_CONTAINER                      
                                       
# reference: https://github.com/jupyter/docker-stacks/blob/975e7404e1b16091fee6ea6795f59104043c8092/tensorflow-notebook/Dockerfile
LABEL maintainer="Jupyter Project <jupyter@googlegroups.com>"
SHELL ["/bin/bash", "-o", "pipefail", "-c"]
                                       
RUN pip config set global.index-url http://mirrors.aliyun.com/pypi/simple
RUN pip config set install.trusted-host mirrors.aliyun.com
                                       
# Install Tensorflow                   
RUN pip --no-cache-dir install tensorflow==2.12.0
                                       
# Install Pytorch                      
RUN pip --no-cache-dir install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
                                       
# Elyra                                
RUN pip install --no-cache-dir --upgrade "elyra[all]>=3.7.0"
                                       
# jupyterlab-lsp                       
RUN pip install --no-cache-dir 'jupyterlab>=3.0.0,<4.0.0a0' jupyterlab-lsp
Run pip install --no-cache-dir -U jedi-language-server
Run pip install --no-cache-dir 'python-lsp-server[all]'
                                       
# lckr-jupyterlab-variableinspector       
RUN pip install --no-cache-dir lckr-jupyterlab-variableinspector
                                       
# HDF5                                 
RUN pip install --no-cache-dir jupyterlab_hdf
RUN jupyter labextension install @jupyterlab/hdf5
                                       
# Some other extensions                
RUN jupyter labextension install jupyterlab-spreadsheet
RUN pip install --no-cache-dir jupyterlab-spellchecker
RUN pip install --no-cache-dir jupyterlab-system-monitor
RUN pip install --no-cache-dir jupyterlab_execute_time
RUN pip install --no-cache-dir 'ipydrawio[all]'
                                       
RUN fix-permissions "${CONDA_DIR}" && fix-permissions "/home/${NB_USER}"                                                                      
                                       
USER root                              
RUN apt update -y                      
RUN apt install vim -y                 
                                       
USER jovyan                            
COPY .pip /home/jovyan                 
WORKDIR /home/jovyan/work

The Dockerfile above was built as follows:

docker build . -t jupyter/tensorflow-2.12_pytorch-2.01:v0.1

Finally I run the docker in command line:

docker run -it --gpus all --rm --name jupyter -p 8888:8888 \ 
            -v /home/zhangxf/workdir:/home/jovyan/workdir \ 
            jupyter/tensorflow-2.12_pytorch-2.01:v0.1

Why tensorflow failed to detector GPU? How can I fix this? Any help would be greatly appreciated!.