GitHub has presented a limited public beta version of GitHub Models. This gives developers free access to various AI models in an interactive sandbox test environment. These can be tested out with a view to introducing them to development environments.
Microsoft’s development platform GitHub recently introduced a limited beta of GitHub Models. The platform gives them access to various LLMs in a sandbox environment to test them securely and eventually use them in production. The offering includes both open-source options such as Llama 3.1 and closed-source models such as OpenAI’s GPT-4o.
Easy access
The goal of the platform is to give all developers the opportunity to gain easier access to these frequently used LLMs. According to GitHub, many developers often do not yet have access to these models. This is despite their increasing desire to build GenAI applications based on a full stack of backend and front-end code and underlying LLMs.
Through GitHub Models, developers can access LLMs such as those from Meta, Cohere, A121 Labs, Mistral AI, Azure OpenAI Service and Microsoft. Offerings include Meta’s Llama 3.1, OpenAI’s GPT-4o and GPT-4o mini, Cohere Command, A121 Jamba, Mistral Large 2 and Phi 3 medium from Microsoft.
Features
Developers can deploy these LLMs via a built-in “playground,” test various prompts and model parameters, and eventually integrate them into development environments. Compatible IDEs include GitHub Codespaces or Visual Studio Code from Microsoft. They can also use GitHub Copilot in these IDEs.
In this way, it is possible, for example, to discover how certain LLMs work in new applications or which LLM is best suited for this purpose. They can also “play” with various integration techniques, such as RAG. All this can be supported with cloud resources. GitHub Codespaces helps here with the necessary inference that local machines often cannot handle and cloud access is too expensive.
Registration for the public beta of GitHub Models is now available through the site. The development platform will be expanded to include other models in the near future, before it becomes generally available.
Also read: GitHub Copilot Workspace: will AI replace software engineers?