temper-ai/Temper-1-0.5B

Temper-1-0.5B: A Specialized Rust Code Generation Model

Temper-1-0.5B is a compact yet powerful 0.5 billion parameter model, fine-tuned from Qwen2.5-Coder-0.5B-Instruct, with a 32768 token context length. Its primary focus is generating Rust code, demonstrating exceptional performance for its size.

Key Capabilities

• Superior Rust Code Generation: Outperforms models up to 1.5B parameters on pass@1 metrics across internal trivial, easy, and medium Rust task sets, as well as HumanEval-RS and MBPP-RS benchmarks.
• Reinforcement Learning Enhanced: Trained on 10,000 compiler-verified Rust examples and further refined using reinforcement learning on its own generated answers.
• Efficient for Rust Development: Designed to solve more Rust tasks on the first attempt compared to larger models, making it highly efficient for Rust-specific coding challenges.

Good For

• Rust Developers: Ideal for generating Rust functions and solving coding problems in Rust.
• Resource-Constrained Environments: Its small size (0.5B parameters) makes it suitable for applications where computational resources are limited, while still delivering strong performance for Rust tasks.
• Chat-based Code Generation: Expects chat requests and provides code within closed code blocks, suitable for interactive coding assistance.

Limitations

• Rust-Specific: Primarily trained on Rust tasks with English requests; may produce irrelevant or made-up answers for other languages or non-Rust related queries.
• No Fill-in-the-Middle: Not designed for inline code completion in editors.
• Async/Web Frameworks: Limited training data on async code and web frameworks, potentially impacting performance in these areas.