What to know

  • JetBrains has released Mellum, its first open AI model for coding, with 4 billion parameters.
  • Mellum is designed for code completion, focusing on speed, accuracy, and efficiency.
  • The model is multilingual and available for public use, including on Hugging Face.
  • Mellum offers solid performance for its size, aiming to make AI coding tools more accessible.

JetBrains, known for its popular development tools, has introduced Mellum, an open-source AI model built specifically for code completion. Mellum is JetBrains' first open AI model for coding and is now available to the public, including through the Hugging Face platform.

Mellum uses a LLaMA-style architecture and contains 4 billion parameters. This makes it efficient for both cloud-based and local deployment. The model is multilingual, allowing it to assist with code in several programming languages. JetBrains describes Mellum as a "focal model," meaning it is specialized for a single task—in this case, code completion—rather than trying to handle a wide range of functions like larger, general-purpose AI models.

This focus on a specific task brings several advantages. Mellum is designed to be faster and more accurate for code completion within JetBrains IDEs. It also aims to lower operational costs and reduce environmental impact. By making the model open and accessible, JetBrains hopes to support researchers and smaller teams who may not have the resources to use larger, more general AI models.

Benchmark data shared by JetBrains shows that Mellum performs well compared to other code models of similar or larger size. While it does not outperform the largest models in every test, Mellum offers solid results for its size, especially in speed and efficiency. This makes it a practical choice for developers who need reliable code completion without the heavy resource requirements of bigger models.

JetBrains plans to continue developing Mellum and expand it into a family of specialized models for different coding tasks. For now, Mellum represents a step forward in making advanced AI coding tools more accessible to a wider range of users.

Via: Tech Crunch