IFMedTechdemo/qwen3-vl-cadquery
The IFMedTechdemo/qwen3-vl-cadquery is an 8 billion parameter vision-language model, fine-tuned from unsloth/Qwen3-VL-8B-Instruct-unsloth-bnb-4bit. Developed by IFMedTechdemo, this model specializes in understanding and generating CadQuery-style Python code for parametric CAD modeling. It is optimized for tasks such as creating CAD code from natural language instructions, modifying existing snippets, and basic geometric reasoning, leveraging a 32768 token context length.
Loading preview...
Model Overview
This model, IFMedTechdemo/qwen3-vl-cadquery, is a fine-tuned Qwen3-VL variant with 8 billion parameters, specifically adapted for CadQuery-style CAD/code understanding and generation. It was developed by IFMedTechdemo and is based on unsloth/Qwen3-VL-8B-Instruct-unsloth-bnb-4bit, licensed under Apache-2.0. The model was trained for 30 epochs on approximately 2000 examples from the ThomasTheMaker/cadquery dataset, utilizing Unsloth and Hugging Face TRL for efficient fine-tuning.
Key Capabilities
- Generates CadQuery-style Python code from natural-language CAD instructions.
- Explains and modifies existing CadQuery snippets, such as adding features, changing dimensions, or applying operations.
- Performs basic reasoning about geometric intent described in text.
Intended Use Cases
This model is designed to assist with CAD and parametric modeling workflows, particularly those involving CadQuery. It can be used for:
- Automating the creation of initial CAD code from design specifications.
- Streamlining modifications to existing CadQuery designs.
- Facilitating the understanding of geometric descriptions in text.
Limitations
Users should be aware that the model may produce code with:
- Syntax errors, import issues, or missing variables.
- Invalid CadQuery API usage.
- Inconsistent dimensions or constraints, requiring validation for production use.
Complex assemblies and tight tolerance requirements may still necessitate human review and iterative refinement.