JetBrains releases Mellum, an ‘open’ AI coding model
JetBrains, the innovative force behind several popular development tools, has unveiled its first openly available AI model for code generation. This new model, called Mellum, is now accessible via the popular AI development platform Hugging Face.
Overview of Mellum
Mellum is designed to enhance coding productivity by offering smart code completions that intuitively pick up on surrounding context. Trained on over 4 trillion tokens and boasting 4 billion parameters, the model is optimized for integration into professional development tools and intelligent coding assistants.
Key Facts About Mellum
- Developed by JetBrains, known for top-tier software development tools.
- Trained on extensive datasets including permissively licensed GitHub code and English-language Wikipedia articles.
- Utilizes around 4 trillion tokens, equivalent to roughly 120,000 lines of code per million tokens.
- Powered by a cluster of 256 H200 Nvidia GPUs over a period of 20 days.
- Released under an Apache 2.0 license and available for further exploration and fine-tuning.
Read also :
Wikipedia says it will use AI, but not to replace human volunteers
Applications and Limitations
Mellum is primarily aimed at enhancing developer tooling, such as augmenting code suggestions in integrated development environments. JetBrains emphasizes that while Mellum is a breakthrough in AI-powered coding, it is not immediately ready for production without further fine-tuning. They have already provided some versions optimized for Python, intended more as a gauge of the model’s potential than as a complete solution.
It is important to note that, like many AI systems trained on publicly available datasets, Mellum may mirror certain biases or styles found in open source code. Developers should remain cautious, as the suggestions generated might not always be secure or free from vulnerabilities.
Looking Ahead
In a recent blog post, JetBrains expressed optimism about Mellum’s future. The company stated, “This is just the beginning. 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 forward-looking approach signals exciting possibilities for both professional developers and researchers in the field of AI-assisted programming.
Final Thoughts
With Mellum, JetBrains reaffirms its commitment to innovation by pushing the boundaries of how code is written and understood. As AI-generated code and tools become more prevalent, it will be crucial for organizations to balance progress with security and reliability. Mellum represents a promising step in that direction.
Read also :
Pinterest launches new tools to fight AI slop
