What is a compute in Azure ML
Compute instances in Azure ML Studio are single-owned cloud-based workstations used as a pre configured, optimized and fully customizable environment for ML applications. They also allow us to share files between instances and built manageable applications.
How to create a compute in Azure ML
Setting a compute instance is a first step prior to development. Typically for ML applications, we must ensure our compute has GPU in order to enable CUDA. The steps to create a compute instance are:
- Create Azure ML Compute : name : Choose an option with GPU
- Select from all options: Standard_NC8as_T4_v3
The are several compute options in Azure ML Studio, the selected one has a reasonable price at 0.75/hr.
What is CUDA and why do we need it?
CUDA is a parallel computing platform and programming model created by NVIDIA used to speed up applications by harnessing the power of GPUs
Do I have CUDA?
CUDA is a standard feature in all NVIDIA GeForce, Quadro, and Tesla GPUs as well as NVIDIA GRID solutions, a comprehensive list of products with CUDA are found here. The command torch.cuda.is_available() is a simple way to check if we have access to GPUs in our environment, it returns True if the system has the NVIDIA driver correctly installed.
Setting up the environment
To create an environment within our compute, we run some steps in the Terminal. A common encountered issue while setting an environment with CUDA is the incompatibility between torch and torchvision packages; the pair torch=2.0.0+cu118 and torchvision=0.15.1 is our choice. After the installation of the previous packages, CUDA should be enabled in our environment.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
conda create -n myenv python==3.8.10
pip install cython
pip install torch==2.0.0+cu118 torchvision==0.15.1+cu118 --extra-index-url https://download.pytorch.org/whl/cu118
conda activate myenv
import torch
torch.cuda.is_available()
pip install opencv-python
pip install azureml
pip install azureml.core
pip install azureml-dataset-runtime --upgrade
conda install pip
conda install ipykernel
python -m ipykernel install --user --name my_kernel --display-name "my_kernel_display"
conda list --explicit > spec.txt
conda create --name myenv --file spec-file.txt
This post showed how to create a compute in azure ML Studio, and do a basic set up for ML applications enabling CUDA in our environment.