---
license: mit
library_name: gguf
base_model: Qwen/Qwen2.5-Coder-7B
datasets:
- AI-MO/NuminaMath-TIR
tags:
- mathematics
- geogebra
- 3d-visualization
- education
- coding
- reasoning
- uvicorn
- fastapi
---

<p align="center">
  <img src="logo.png" alt="ΣMath Visual Core v2.0 Logo" width="550"/>
</p>

# ΣMath — Visual Computation Engine v2.0

### **Powered by Qwen2.5-Coder-7B & NuminaMath-TIR**

**Developed by: Khurram Pervez, Assistant Professor of Mathematics**

**ΣMath Core** is a high-performance mathematical visualization engine that bridges the gap between deep symbolic reasoning and real-time interactive rendering. By leveraging a fine-tuned **Qwen2.5-Coder-7B** backbone with the **NuminaMath-TIR** dataset, the model excels at **Chain-of-Thought (CoT)** reasoning, allowing it to solve complex geometric problems before translating them into interactive code.

The engine utilizes a specialized **Resilient Execution Pipeline** to render 3D manifolds, animations, and parametric surfaces directly in the browser, optimized specifically for local deployment on NVIDIA hardware.

## 🚀 The Multi-Stage Pipeline

### 1. TIR (Thought-Intermediate-Reasoning)
By training on the **NuminaMath-TIR** dataset, the model follows a rigorous logical path:
* **Identification:** Analyzes the geometric properties of the requested manifold.
* **Calculation:** Determines the necessary vertices, normals, and parametric equations.
* **Code Synthesis:** Generates high-efficiency Python code (Plotly/Matplotlib) using its native **Coder** capabilities.

### 2. The Resilient Engine (FastAPI Layer)
To ensure stability during research, the system includes a proprietary processing layer:
* **Dummy Interception:** Captures and silences `plt.show()` commands to prevent GUI thread blocking on Ubuntu/Linux servers.
* **Colorscale Transpilation:** Automatically maps Matplotlib colormap names (e.g., *spring, summer*) to Plotly-valid equivalents to ensure 3D renders never fail.
* **Sandbox Execution:** Executes generated code in a safe local scope using your **RTX 4060 Ti**.

## 📸 Interactive Visual Samples

Here are examples of advanced parametric surfaces generated in real-time by **ΣMath Core v2.0**, showcasing the full **Thought-Intermediate-Reasoning (TIR)** pipeline.

| 3D Torus Visualization | Full Research Dashboard Interface | Resilient Color Scaling Error Fix |
| :---: | :---: | :---: |
| <img src="viz.png" alt="ΣMath Interactive Torus" width="100%"/> | <img src="dashboard.png" alt="ΣMath Dashboard" width="100%"/> | <img src="fix.png" alt="Resilient Colorscale Error" width="100%"/> |

## 💻 System Configuration

| Component | Specification |
| :--- | :--- |
| **Compute Engine** | NVIDIA GeForce RTX 4060 Ti (16GB VRAM) |
| **Model Format** | GGUF (Quantized Q4_K_M) |
| **Context Window** | n_ctx=4096 (Optimized for detailed manifold calculation) |
| **OS** | Ubuntu 22.04 LTS (Optimized for `Agg` Backend) |
| **Frameworks** | FastAPI, Llama-cpp-python, Plotly, mpld3 |

## 🛠️ Quick Start

### 1. Installation
```bash
# Clone this repository
git clone [https://huggingface.co/Khurram123/SigmaMath-Visual-Core](https://huggingface.co/Khurram123/SigmaMath-Visual-Core)
cd SigmaMath-Visual-Core

# Install dependencies
pip install fastapi uvicorn llama-cpp-python numpy matplotlib mpld3 plotly