news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Machine Heart Report

Editor: Du Wei, Jia Qi

GitHub's new feature "GitHub Models" is expected to accelerate the arrival of the era of AI engineers.

What? The familiar code hosting platform GitHub has evolved again! The platform has also begun to provide Playground for large AI models.

All the popular large models in the industry that you can name, including Microsoft's Phi-3, OpenAI's GPT-4o, Meta's Llama 3.1, Cohere's Command R+, and Mistral AI's Mistral Large, can be tried in an interactive sandbox.

In the coming months, Github will also add more language, vision, and other types of models.

In other words, the models in this picture can all be obtained for free! This is equivalent to a new way to test various large models for free!

Not only that, developers can also easily import suitable models into their own AI projects. Github has created a fast channel that can directly place models in Codespaces and VS Code development environments, which has lowered the threshold for deploying AI models.

This is the “GitHub Models” feature launched by GitHub today!

For developers, as long as they have the right tools and training, everyone can become an AI engineer. Developers select models, code in GitHub codespaces, and then deploy them in production through Azure, providing a "one-stop" service.

Specifically, you can call the GitHub Models command in the GitHub CLI, use it in GitHub Actions through a series of JSON files, and use GitHub Models to build GitHub Copilot extensions to cover the entire cycle of application development.

When the project entered the online stage, Github also provided seamless integration with Azure AI. Github's authentication could be subscribed through Azure, thus lowering the threshold for deploying AI applications to the production environment.

Now, you can deploy AI applications in more than 25 Azure regions around the world through Github and get Azure's enterprise-level security protection.

Consistent with GitHub and Microsoft's ongoing commitment to privacy and security, any hints or outputs from GitHub models are not shared with model contributors or used to train or improve the models.

Of course, trying out "GitHub Models" is not completely unrestricted. Individual users are limited to 150 accesses per day and no more than 15 per minute. Each request can process up to 8,000 tokens and output up to 4,000 tokens.

In this regard, GitHub CEO Thomas Dohmke said that the launch of this feature marks another transformation of GitHub, from creating AI through open source collaboration to creating software with the power of AI, and now using GitHub Models to promote the rise of AI engineers.

And based on OSS repositories, Copilot Extensions, and GitHub Models features, GitHub hopes to bring as many partners as possible to its platform.

A developer demonstrated a case of running the model directly in the GitHub code space. There is no need to install anything and it can be started within a few seconds.

Image source: https://x.com/DanWahlin/status/1819113874689610133

I wonder how Hugging Face will respond when Github becomes “Hugging Face”?