intel / ai-containers Goto Github PK

Describe the bug

I was trying to use the images documented here. However, they do not appear to have Jupyter in them:

docker run -it --rm \
    -p 8888:8888 \
    --device /dev/dri \
    -v /dev/dri/by-path:/dev/dri/by-path \
    --ipc=host \
    intel/intel-extension-for-pytorch:2.1.20-xpu-pip-base
(idp) root@5711063c540f:/# python -m jupyter --version
/opt/conda/envs/idp/bin/python: No module named jupyter

Within the container, if I manually install jupyter then the server is there:

(idp) root@5711063c540f:/# cat << EOF > requirements.txt
jupyterlab==4.3.0a0
jupyterhub==5.0.0
notebook==7.3.0a0
jupyter-server-proxy>=4.1.2
EOF
(idp) root@5711063c540f:/# pip install -r requirements.txt
(idp) root@5711063c540f:/# python -m jupyter --version
Selected Jupyter core packages...
IPython          : 8.26.0
ipykernel        : 6.29.5
ipywidgets       : not installed
jupyter_client   : 8.6.2
jupyter_core     : 5.7.2
jupyter_server   : 2.14.1
jupyterlab       : 4.3.0a0
nbclient         : 0.10.0
nbconvert        : 7.16.4
nbformat         : 5.10.4
notebook         : 7.3.0a0
qtconsole        : not installed
traitlets        : 5.14.3

Error Logs

/opt/conda/envs/idp/bin/python: No module named jupyter

Reproduction Instructions

docker run -it --rm \
    -p 8888:8888\
    --device /dev/dri \
    -v /dev/dri/by-path:/dev/dri/by-path \
    --ipc=host \
    intel/intel-extension-for-pytorch:2.1.20-xpu-pip-base \
    python -m jupyter --version
/opt/conda/envs/idp/bin/python: No module named jupyter

I have tried the tags xpu-jupyter, 2.1.20-xpu-pip-base, and 2.1.10-xpu-pip-base all with the same effect.

### Affected Subfolder

- [ ] classical-ml
- [ ] enterprise
- [ ] preset
- [ ] python
- [X] pytorch
- [ ] tensorflow
- [ ] test-runner
- [ ] workflows

### Versions

```Markdown
Using the image published at intel/intel-extension-for-pytorch:{xpu-jupyter,2.1.20-xpu-pip-base,2.1.10-xpu-pip-base}

Tracking issue for:

• https://github.com/intel/ai-containers/security/code-scanning/682

Description

Is there a version of this image with that uses AVX512 available?

https://hub.docker.com/r/intel/intel-optimized-tensorflow-avx512 seems unmaintained, so I'm hoping these images can fill that gap :)

tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: AVX2 AVX512F AVX512_VNNI AVX512_BF16 AVX_VNNI AMX_TILE AMX_INT8 AMX_BF16 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.

Describe the solution you'd like

Duplicate all tags of https://hub.docker.com/r/intel/intel-extension-for-tensorflow to have a version that uses AVX512.

Describe alternatives you've considered

No response

Scorecard wants a hash, but I would be alright with just a pinned version :)

Tracking issue for:

• https://github.com/intel/ai-containers/security/code-scanning/696

As @ma-pineda pointed out, sometimes CI will fail to build these groups because GitHub can only guarantee 14GB of storage, but these images are more like 20-25 before compression.

As a workaround:

sudo rm -rf /usr/share/dotnet
sudo rm -rf /opt/ghc
sudo rm -rf "/usr/local/share/boost"
sudo rm -rf "$AGENT_TOOLSDIRECTORY"

Or switch to a larger runner.

Tracking issue for:

• https://github.com/intel/ai-containers/security/code-scanning/1643
• https://github.com/intel/ai-containers/security/code-scanning/1644
• https://github.com/intel/ai-containers/security/code-scanning/1645

Describe the bug

While trying to add some additional functionality to a derived version of the ipex-xpu-jupyter container, attempting to access the shell results in bash complaining that conda does not exist:

docker run -it --rm \
    -p 8888:8888 \
    --net=host \
    --device /dev/dri \
    -v /dev/dri/by-path:/dev/dri/by-path \
    --ipc=host \
    intel/intel-extension-for-pytorch:2.1.20-xpu-pip-jupyter \
    /bin/bash
WARNING: Published ports are discarded when using host network mode
bash: conda: command not found

Conda is not installed in the intel-xpu-base layer, nor intel-xpu-jupyter layer. However, the Dockerfile layer for intel-xpu-jupyter is adding 'conda activate idp' regardless. This is due to this check in the Dockerfile:

RUN if [ ! -d "$(which conda)" ]; then \
    echo "conda activate idp" >> ~/.bashrc; \
    fi

That check is attempting to check if conda exists, and then add the conda activate idp to the .bashrc.

However, which returns the path to the executable (if it is in the path) and not a file. The the '-d' check will fail if conda is installed (which returns file path) and if conda is not installed (which returns nothing, so no directory). The '!' inverts that failure to be success. The net result is the conda command will be added to ~/.bashrc regardless of whether conda exists or not.

A corrected line:

# If conda is in the path, activate idp environment
RUN if $(which conda >/dev/null); then \
    echo "conda activate idp" >> ~/.bashrc \
fi

The instructions are also telling the user to publish a port (-p) and use --net=host; you shouldn't need --net=host since you are publishing the internal port 8888 to the host port 8888.

Error Logs

WARNING: Published ports are discarded when using host network mode
bash: conda: command not found

Reproduction Instructions

docker run -it --rm \
    -p 8888:8888 \
    --net=host \
    --device /dev/dri \
    -v /dev/dri/by-path:/dev/dri/by-path \
    --ipc=host \
    intel/intel-extension-for-pytorch:2.1.20-xpu-pip-jupyter \
    /bin/bash

Affected Subfolder

• classical-ml
• enterprise
• preset
• python
• pytorch
• tensorflow
• test-runner
• workflows

Versions

intel/intel-extension-for-pytorch:2.1.20-xpu-pip-jupyter

Describe the bug

The package intelpython3_full fails when installing version 2024.2.0. Switching back to 2024.1.0 does not cause this issue, and it appears that the latest version has been added/updated 5 days ago: https://anaconda.org/intel/intelpython3_full/files

Error Logs

=> ERROR [idp idp 4/6] RUN conda run -n idp python -m pip install --no-cache-dir -r idp-requirements.txt                                                                                                                        0.7s
------
 > [idp idp 4/6] RUN conda run -n idp python -m pip install --no-cache-dir -r idp-requirements.txt:
0.590 
0.590 EnvironmentLocationNotFound: Not a conda environment: /opt/conda/envs/idp/envs/idp
0.590 
------

### Remove the failing command

idp-1  | # conda environments:
idp-1  | #
idp-1  |                          /opt/conda
idp-1  | base                     /opt/conda/envs/idp
idp-1 exited with code 0

### This output is incorrect, it should look like

idp-1  | # conda environments:
idp-1  | #
idp-1  | base                     /opt/conda
idp-1  | idp                      /opt/conda/envs/idp
idp-1  | 
idp-1 exited with code 0

Reproduction Instructions

cd python
IDP_VERSION=full docker compose up --build
# comment out lines 75-79
IDP_VERSION=full docker compose up --build
# switch back to last version, uncomment lines 75-79
IDP_VERSION=full==2024.1.0 docker compose up --build

Affected Subfolder

• classical-ml
• enterprise
• preset
• python
• pytorch
• tensorflow
• test-runner

Versions

Docker Compose version v2.26.1