electrocampbell/nebula-8lang-14b
The electrocampbell/nebula-8lang-14b is a 14 billion parameter language model fine-tuned by electrocampbell from Qwen/Qwen2.5-14B. It specializes in translating code from Nebula, a universal intermediate language, into 8 target programming languages including Python, JavaScript, and Rust. This model is optimized for code translation tasks, demonstrating strong performance in converting Nebula code to idiomatic target languages.
Loading preview...
Model Overview
The electrocampbell/nebula-8lang-14b is a 14 billion parameter model, fine-tuned from the Qwen/Qwen2.5-14B base model using LoRA (SFT) with a rank of 16. Developed by electrocampbell, its primary function is to translate code from Nebula, a token-efficient universal intermediate language, into 8 specific programming languages: Python, JavaScript, TypeScript, Go, Swift, Kotlin, Rust, and C.
Key Capabilities
- Nebula to 8-Language Translation: Excels at converting Nebula intermediate code into idiomatic code for the specified target languages.
- Code Generation: Facilitates the generation of functional code in multiple languages from a common intermediate representation.
- Performance: Achieves a raw Pass@1 score of 89.0% on HumanEval (Nebula→Python) when run on optimal hardware (2x H100 80GB), significantly outperforming its 7B counterpart.
Training and Data
The model was trained for 3 epochs on the electrocampbell/nebula-8lang-203k dataset, comprising 203,000 code translation pairs. This specialized dataset ensures its proficiency in the targeted translation tasks.
Ideal Use Cases
- Cross-Language Code Porting: Automating the conversion of code logic between different programming languages via the Nebula intermediate form.
- Code Generation Tools: Integrating into development tools that require generating code in various languages from a unified source.
- Educational Platforms: Assisting in understanding how code logic translates across different language syntaxes.