Model Overview

Etherll/Mellum-4b-sft-rust is a 4 billion parameter large language model (LLM) built upon the JetBrains/Mellum-4b-base architecture, which itself was pre-trained on over 4 trillion tokens. This model has undergone specialized fine-tuning to excel in Rust code Fill-in-the-Middle (FIM) tasks, making it highly proficient in completing Rust code snippets based on context.

Key Capabilities

• Specialized for Rust FIM: Optimized specifically for accurately filling in missing parts of Rust code.
• Robust Base Model: Benefits from the strong foundation of JetBrains' Mellum-4b-base model.
• Efficient Deployment: Designed to be efficient for both cloud and local inference environments.
• IDE Integration Ready: Particularly effective when integrated with developer tools like Continue.dev for an enhanced coding assistant experience.
• GGUF Version Available: A GGUF version is provided for local CPU inference using tools such as llama.cpp or Ollama.

Fine-tuning Details

The model was fine-tuned on the Etherll/CodeFIM-Rust-Mellum dataset, comprising approximately 57,000 Rust-specific FIM examples. It expects a specific FIM input format:

<filename>{{{filename}}}
<fim_suffix>{{{suffix_code}}}<fim_prefix>{{{prefix_code}}}<fim_middle>

Good For

• Developers working with Rust who need intelligent code completion.
• Integrating into IDEs or custom developer tooling for Rust FIM functionality.
• Local inference on CPU or GPU for Rust code assistance.