JetBrains releases Mellum, an open AI coding model
JetBrains, known for its powerful app development tools, has taken a significant step by launching its first “open” AI model for coding. This new model, called Mellum, marks the company’s efforts to blend artificial intelligence with professional coding expertise.
JetBrains releases Mellum, an open AI coding model : An Overview of Mellum
Mellum is designed to enhance code completion by suggesting code snippets based on surrounding context. As an open AI coding model, it leverages its understanding of programming patterns to deliver accurate and context-aware suggestions. With a model built on more than 4 trillion tokens and 4 billion parameters, this open AI coding model offers a strong problem-solving capability. In this context, tokens represent bits of raw data processed by the model—an open AI coding model that learns from vast codebases—where a million tokens roughly correspond to about 30,000 lines of code. Developers benefit from the open AI coding model’s ability to interpret context and generate code that aligns with best practices and project requirements.
Key Features and Intended Use
Mellum is built for integration into developer tools and AI-assisted coding environments, helping professionals receive intelligent code suggestions in real-time. It also opens up opportunities for research in code understanding and generation, making it a valuable resource for both educational initiatives and experimental fine-tuning.
Training and Availability
The model has been trained on a compilation of datasets, including permissively licensed code from GitHub and English Wikipedia articles. JetBrains completed the training in about 20 days using a cluster of 256 H200 Nvidia GPUs. Moreover, Mellum is available under the Apache 2.0 license and can be accessed on the AI development platform Hugging Face. To learn more about the model, visit the official page on Hugging Face.
Fine-Tuning for Production
It’s important to note that Mellum, in its base form, requires fine-tuning to be used effectively. Although JetBrains has shared several versions of Mellum fine-tuned for Python, they advise against deploying these versions directly into production, suggesting they are better for initial assessments and experiments.
Read also :Â
Startups launch products to catch people using AI cheating app Cluely
Addressing Security Concerns
As AI-generated code becomes more prevalent, potential security risks also emerge. Recent research indicates that over 50% of organizations sometimes or frequently face security issues with AI-produced code. This highlights the need for careful evaluation when integrating such models into critical systems. An informative report from developer security platform Snyk offers more details on these challenges, which you can review by following this link.
The Road Ahead
JetBrains cautions that Mellum might mirror biases from the public codebases it was trained on, and its suggestions are not guaranteed to be completely secure or devoid of vulnerabilities. In a recent blog post, the company emphasized that this launch is just the beginning. They are focused on building a model with a specific purpose and hope that Mellum sparks innovative experiments, collaborations, and meaningful contributions in the coding community.
With Mellum now available, developers have a new tool to explore the potential of AI in coding, paving the way for more specialized and effective solutions in software development.
Read also :Â
Tesla board member Joe Gebbia bought some stock
