From the course: GitHub Administration Cert Prep
Key concepts of the GitHub ecosystem - GitHub Tutorial
From the course: GitHub Administration Cert Prep
Key concepts of the GitHub ecosystem
- [Instructor] Let's talk a little bit about GitHub for teaching MLOPs, and this applies to professionals as well as students. First, we have reproducibility. The Codespaces environment allows you to get this incredible amount of reproducibility so that other people on your team or other students in the classroom have access to the same environment that you do. Next, when you dive into a GPU, this allows you to use the GPU to do things like use pre-trained models or to fine tune a model or to use modern AI tools like Open AI's Whisper. The AI coding assistant, Copilot is invaluable because allows you to prompt and ask questions back and forth with the Copilot system and actually get suggestions about how to write your code, how to build your code, how to build tools, and potentially even build the boilerplate code for you. With continuous integration and deployment via GitHub actions, it's an amazing way to deploy your application in the same location that you built it in. Now, reproducibility with Codespaces is all about three things. It's a cloud-based development environment and a workspace that has a container image built in. This could be a Ubuntu image or a Microsoft created image. It could be any image that works with Docker, and then you could customize it to meet your own needs through your own configurations with .devcontainer. Also, with compute and storage, you can actually customize Codespaces to use GPUs or to use multi-core machines or to use high memory machines. And this is invaluable because you can test and build in a similar environment as production. Access to the GPU allows you to do things like dive into Hugging Face, one of the leading vendors of pre-trained models. You can take that pre-trained model down, fine tune it, and actually do it right on your device without having to pay for a $5,000 GPU. Hugging Face works very frequently with PyTorch. You have access to PyTorch. You also can get into TensorFlow, build your own deploy models from scratch, or even go at a lower level. Dive into the Nvidia Cuda SDK and build functions that get applied directly to the GPU. The Copilot AI Coding Assistant is a way of getting suggestions, adding details directly into your project, and with the Copilot feeding everything inside of your project, you can then ask for linting from PyLint or testing from PyTest or execution with IPython, and get this virtuous cycle of continuously improvement. And this really can be a three or four times X improvement and speed. Now finally, with GitHub actions, continuous integration and continuous delivery, it allows you to use this codespace, do a make install step, do a make test step, do a make deploy step inside of your make file, then push that into your continuous integration steps with GitHub actions, and then that can trigger the push of an image to Amazon's container registry. That would then in turn, go and trigger the platform as a service offering App Runner, which would in turn deploy your application. So you have all the tools necessary to develop in the exact same environment that you're going to eventually push deployment to. And it makes for a smooth transition for continuous integration and continuous delivery best practices.
Download courses and learn on the go
Watch courses on your mobile device without an internet connection. Download courses using your iOS or Android LinkedIn Learning app.