JetBrains, a well-known provider of software development tools, has released its first "open" AI model for coding, named Mellum. On Wednesday, the company made Mellum publicly available on Hugging Face, a prominent AI development platform. This model, initially launched for JetBrains' software development suites last year, is now accessible to a broader audience, marking a significant milestone in the company's AI initiatives. Mellum is a code-generating model trained on over 4 trillion tokens and boasts 4 billion parameters. These parameters are akin to the problem-solving abilities of the model, while tokens represent the raw data processed during training. For context, a million tokens roughly equate to about 30,000 lines of code. According to JetBrains, Mellum is designed for integration into professional developer tooling, such as intelligent code suggestions in integrated development environments (IDEs), AI-powered coding assistants, and research on code understanding and generation. It is also suitable for educational applications and fine-tuning experiments. The model is released under the Apache 2.0 license, ensuring that developers and researchers can use, modify, and distribute it freely. Mellum was trained on a variety of datasets, including permissively licensed code from GitHub and English-language Wikipedia articles. The training process took approximately 20 days using a cluster of 256 H200 Nvidia GPUs, showcasing the substantial computational resources required for such an endeavor. However, using Mellum is not as simple as downloading and deploying it immediately. The base model requires fine-tuning before it can be effectively utilized. JetBrains has provided a few pre-fine-tuned models for Python, intended to demonstrate Mellum's potential capabilities. The company explicitly warns that these models are not ready for production environments and are meant for evaluation purposes only. While AI-generated code has the potential to revolutionize software development, it also introduces new security challenges. According to a late 2023 survey by developer security platform Snyk, more than 50% of organizations experience security issues with AI-produced code, either sometimes or frequently. JetBrains acknowledges these concerns, noting that Mellum's code suggestions may reflect biases present in public codebases and might not be secure or free from vulnerabilities. In their blog post, JetBrains emphasized that this release is just the beginning of their journey in AI coding assistance. They stated, "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 announcement highlights JetBrains' commitment to advancing AI in the context of software development while fostering a collaborative and transparent research community.