Lumorux 1.0 GGUF
quantized model for coding and 3D graphics
2 min readupdated
This model is a 4-bit GGUF version (format for running on consumer hardware) of the Qwen2.5-Coder-7B-Instruct fine-tune. It is trained to help developers and 3D artists with coding, web development, and creating scripts for Blender (via Python API) and Three.js / WebGL. It differs from the base Qwen model through its narrow specialization in programming and 3D packages. A useful model for local execution on a regular PC or laptop with low VRAM (video memory).
Key features
- Based on the efficient Qwen2.5-Coder-7B-Instruct architecture with 7.6 billion parameters.
- 4-bit Q4_K_M quantization ensures a compact file size (~4.7–5.0 GB).
- Further trained on datasets for Blender Python API, WebGL, Three.js, and web development.
- Supports context up to 4096 tokens.
What you can do
- Write and debug code in Python, JavaScript, C++, Rust, and other languages locally via llama.cpp or Ollama.
- Create automation scripts for Blender through the built-in Python API.
- Generate scenes and shaders for Three.js and WebGL.
Example
Running the model via llama.cpp CLI:
./llama-cli \
-m lumorux-1.0-Q4_K_M.gguf \
--sys-prompt "You are Lumorux, an expert coding and 3D assistant." \
-p "Write a Blender Python script to add a bevel modifier and export as GLB." \
-n 1024 \
--temp 0.6 \
--top-p 0.9 \
-c 4096Read more
DetailsBase model: Qwen/Qwen2.5-Coder-7B-Instruct
Training method: QLoRA (rank 64, alpha 128)
VRAM/RAM requirements: ~5.5–6.5 GB including context
Limitations: not a general-purpose chatbot, optimized for technical and creative code










Comments