ebrown/nlp-gpu-jupyter repository overview

UPDATE:
Version 5.0.0 Updates Python 3.11.7 XGBoost 2.0.3 PyTorch 2.1.2 NumPy + SciPy use MKL now
Version 4.0.1 Update to Tensorflow 2.15. CUDA 12.2 Update supporting packages
Version 3.0.0 Update to Tensorflow 2.14. Update supporting packages
Version 1.6.12 Pick up minor updates. Last release until Tensorflow 2.14
Version 1.6.11 Update to latest packaging version for node.js. Updates to jupyter config, delete and recreate your root volume if you're using one
Version 1.6.10 Update Python 3.11.5; Update Git 2.42.0; Update to Supporting Packages;
Version 1.6.9 Update Supporting Packages; Change TensorRT imports to reduce image size by 3GB by not draggging in dependent development packages.
Version 1.6.8 Supporting Package Updates: JupyterLab updated to 4.0.3; LSP packages removed from JupyterLab.
Version 1.6.7 Added Package hydra-core
Version 1.7.0 Tensorflow 2.13, Remove numba; numpy 1.24.3; scikit-learn 1.3.0 Protobuf errors restoring datasets from disk https://github.com/tensorflow/tensorflow/issues/61351⁠
Version 1.6.5 Update to tensorrt-8.6.1.6-1, minor python package updates
Version 1.6.4 XGBoost 1.7.6
Version 1.6.3 Pin datasets to 2.10.0 to avoid problem with evaluate. Include bluert
Version 1.6.2 Created a link to allow pip to work instead of pip3. Updated supporting packages
Version 1.6.0 and Latest are now Python 3.11.4.

This image is huge (18.5 GB uncompressed) and intended to be a full stack development / execution environment for HPC and desktop applications. This relieves the workload of maintaining different environments on multiple systems. NVIDA GPUs are supported for PyTorch and Tensorflow and XGBoost. The userland is Rocky Linux release 8.7. If you're looking for a one stop image with GPU support and don't have any constraints around image size. This is a good place to start.

This image takes a cue from the published tensorflow jupyter images and expects your local drive to be mounted to /tf/notebooks to be accessed by the jupyter server. If you intend to use this image with vscode or jetbrains gateway and don't want to keep a container, I suggest creating a volume and mounting it to /root to retain the transient configuration add -v docker_volume_name:/root to the commands below in addition to the /tf/notebooks mount. Where docker_volume_name was created with docker volume create docker_volume_name.

docker run -it --rm --gpus all -p 6006:6006 -p 8888:8888 -v "/local/notebook/path:/tf/notebooks" ebrown/nlp-gpu-jupyter:latest

starts jupyterlab

docker run -it --entrypoint=/bin/bash --rm --gpus all -p 6006:6006 -p 8888:8888 -v "/local/notebook/path:/tf/notebooks" ebrown/nlp-gpu-jupyter:latest

runs a shell