How AI describes Mellum

Mellum-4b-base is JetBrains' first open-source large language model optimized for code-related tasks, designed for code completion and intelligent code suggestions in IDEs. It uses a 4-billion-parameter LLaMA-style architecture with an 8,192-token context window and was trained on about 4.2 trillion tokens, enabling efficient cloud inference and local deployment (via vLLM, llama.cpp, or Ollama). The model supports supervised fine-tuning and reinforcement learning adaptation, and is released with a base model plus Python SFT variants for integration into developer tooling and research on code understanding and generation.