Miniconda - Mint Starter Kit

If you need to use other versions of CUDA or cuDNN, you can easily set up the required environment in just a few steps using Miniconda. We recommend choosing the Miniconda image when creating an instance (it does not come with any deep learning frameworks pre-installed, keeping the runtime environment clean and avoiding unnecessary issues). The following example demonstrates how to use Conda to set up an environment for TensorFlow 1.15.0.

If you need to install the virtual environment on a data disk, please refer to the method described at the end of the document.

Creating a Virtual Environment

bash

conda create -n tf python=3.7           # Create a virtual environment named "tf"
conda init bash && source /root/.bashrc # Update the environment variables in bashrc
conda activate tf                       # Switch to the created virtual environment "tf"

Installing Software Dependencies

Here is an example of installing TensorFlow 1.15.0.

bash

# After switching to the Conda virtual environment
conda install tensorflow-gpu==1.15.0
# Conda will automatically resolve and install the required CUDA and cuDNN versions for TensorFlow 1.15.0

python

# Simple test using Python after installation
import tensorflow as tf
hello = tf.constant('Hello, TensorFlow!')
sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))
print(sess.run(hello))

Switching the Notebook Environment

How to Use the New Conda Environment in JupyterLab Notebook

bash

# Create a new Conda virtual environment (skip if already created)
conda create -n tf python=3.7             # Create a virtual environment named "tf"
conda init bash && source /root/.bashrc   # Update the environment variables in bashrc

# Add the new Conda virtual environment to JupyterLab
conda activate tf                         # Switch to the created virtual environment "tf"
conda install ipykernel                    # Install ipykernel to enable kernel integration
ipython kernel install --user --name=tf   # Set up the kernel for the environment, with "tf" as the kernel name

After executing the above commands, if you create a new Notebook, you can select the Notebook named “tf”.

If it is an existing Notebook

Clearing the Conda Virtual Environment

bash

# Clear the installed environment
conda deactivate          # Exit the current (tf) environment and return to the base environment
conda remove -n tf --all  # Remove the tf environment

Deleting Installation Packages and Cache

bash

conda clean -y --all

Installing the Virtual Environment to the Data Disk

Execute the following commands to set up the virtual environment installation path to /root/guphub-tmp/conda/envs and the package cache path to /root/guphub-tmp/conda/pkgs:

bash

mkdir -p /root/gpuhub-tmp/conda/pkgs
conda config --add pkgs_dirs /root/gpuhub-tmp/conda/pkgs

mkdir -p /root/gpuhub-tmp/conda/envs
conda config --add envs_dirs /root/gpuhub-tmp/conda/envs

Verifying the Configuration

bash

# Check if the configured paths are in the file content
cat /root/.condarc

Cancelling the Configuration for Installing Virtual Environment to the Data Disk Edit the /root/.condarc file and delete the lines corresponding to the configured paths.

​Creating a Virtual Environment

​Installing Software Dependencies

​Switching the Notebook Environment

​Clearing the Conda Virtual Environment

​Deleting Installation Packages and Cache

​Installing the Virtual Environment to the Data Disk

Creating a Virtual Environment

Installing Software Dependencies

Switching the Notebook Environment

Clearing the Conda Virtual Environment

Deleting Installation Packages and Cache

Installing the Virtual Environment to the Data Disk