Rhino Coder 7B: Specialized for Rhino3D Python Scripting
Rhino Coder 7B is a fine-tuned version of the Qwen2.5-Coder-7B-Instruct model, specifically engineered to generate accurate Python code for Rhino3D using rhinoscriptsyntax and RhinoCommon APIs. This 7.6 billion parameter model addresses the limitations of general code models by drastically reducing API hallucination and producing more concise, correct scripts for Rhino3D tasks.
Key Capabilities & Differentiators
- Rhino3D API Expertise: Unlike its base model, Rhino Coder 7B correctly utilizes
rhinoscriptsyntax and RhinoCommon APIs, avoiding common hallucinations. - Concise Code Generation: Benchmarks show a significant reduction in average code lines (-3.7) and characters (-40%) compared to the base model for Rhino scripting tasks, leading to less bloated and more efficient code.
- Specialized Training: Fine-tuned using LoRA on a curated dataset of 5,060 instruction-code pairs derived from RhinoCommon API docs, RhinoScriptSyntax source, official samples, and synthetic generation.
- Optimized for MLX: The model is fused and optimized for inference on Apple Silicon via MLX, with usage examples provided for both MLX-LM and as an OpenAI-compatible server.
Ideal Use Cases
- Generating Python scripts for Rhino3D (Rhino 8 REPL workflow).
- Automating computational design and 3D modeling tasks within Rhino.
- Developers needing highly accurate and concise code for Rhino's specific APIs.