<aside> 💡
This page is a simple demo for MLOps, where I will show the process of the whole pipeline from data versioning, processing the data, training the model, and eventually deploying the model. I want to include CI / CD in the future but it’ll depend on my free time.
</aside>
https://github.com/YoKummy/MLOps
Before diving into the project, make sure you have the right environment, with the help of requirement.txt you should be able to do it under python 3.10+. Also, conda is recommended!
We can start by using:
conda create -n mlops python=3.10 -y
This is the first step of the pipeline, where I will introduce DVC(Data Version Control) and show you the fundamental of using dvc and training with YOLO.
<aside> ❓
Why do we need DVC?
Normally in real life scenario, there’s too many data(100k+), github can’t upload so many files, and it’s harder for us to track which data is new, which data is bad, which data is the best performing one, that’s why we need DVC to manage the data while the rest is handled by Github.
</aside>
This is the next step, after we have our data and model ready, it’s time to deploy it, but how? Find out in this page.
<aside> 😃
Now that we have our model and DVC setup, what’s next? We deploy them with fastAPI act as simple demo for future expansion!
</aside>
And now, after we have a working demo to demonstrate our model’s prediction, we need to think about Continuous integration and Continuous deployment
<aside> 💡
Now that we have our model and DVC setup, what’s next? We deploy them with fastAPI act as simple demo for future expansion!
</aside>