## JetBrains Opens the Door to AI Code Generation with Mellum Model
JetBrains, renowned for its suite of developer tools, has taken a significant step into the world of artificial intelligence by releasing Mellum, its “open” AI model designed for coding. The company announced on Wednesday that Mellum, previously utilized within its software development suites, is now openly accessible on the Hugging Face AI platform.
Mellum, a code-generating model trained on a massive dataset of over 4 trillion tokens, boasts 4 billion parameters. Its primary function is code completion, intelligently suggesting code snippets based on the surrounding context. To put that in perspective, a million tokens translates to approximately 30,000 lines of code. Parameters, in simple terms, represent a model’s problem-solving capabilities.
According to a technical report from JetBrains, Mellum is designed for integration into professional developer tools like intelligent code suggestions in integrated development environments (IDEs), AI-powered coding assistants, and for research purposes related to code understanding and generation. The model’s versatility also extends to educational applications and fine-tuning experiments.
JetBrains emphasizes that Mellum, which is licensed under Apache 2.0, was trained using a diverse collection of datasets, including permissively licensed code from GitHub and English-language articles from Wikipedia. The training process itself took around 20 days, utilizing a cluster of 256 H200 Nvidia GPUs.
While Mellum is now available, it’s important to note that it requires fine-tuning before it can be effectively used. The base model isn’t designed for immediate, out-of-the-box deployment. JetBrains has provided some pre-tuned models for Python, but they explicitly state that these are intended for evaluating Mellum’s potential capabilities and not for use in production environments.
The rise of AI-generated code is undoubtedly reshaping the software development landscape, but it also introduces new challenges, particularly in the realm of security. A late 2023 survey by Snyk, a developer security platform, revealed that over 50% of organizations encounter security issues, sometimes or frequently, when using AI-produced code.
JetBrains acknowledges these potential risks, noting that Mellum may inherit biases present in public codebases and that its code suggestions might not always be secure or vulnerability-free.
“This is just the beginning,” JetBrains stated in a blog post accompanying the release. “We’re not chasing generality — we’re building focus. If Mellum sparks even one meaningful experiment, contribution, or collaboration, we would consider it a win.” This release signals JetBrains’ commitment to fostering innovation and collaboration within the AI-assisted coding space, while also highlighting the importance of responsible development and awareness of potential security implications.