adityakum667388/lumichats-v1.2-7b-bnb-4bit
adityakum667388/lumichats-v1.2-7b-bnb-4bit is a specialized 7 billion parameter vision-language model, built on Qwen2.5-VL-7B-Instruct and fine-tuned by adityakum667388 using LoRA. This 4-bit quantized model excels at converting handwritten mathematical formulas from images into accurate LaTeX code. It is optimized for mathematical OCR, providing high accuracy and efficient inference for educational and technical applications.
Loading preview...
Model Overview
LumiChats v1.2 7B is a specialized vision-language model, fine-tuned by adityakum667388 for Image-to-LaTeX OCR. Built upon the Qwen2.5-VL-7B-Instruct base model, it leverages LoRA (Low-Rank Adaptation) for efficient fine-tuning, resulting in a 4-bit quantized model optimized for performance and memory efficiency.
Key Capabilities
- Handwritten Formula Recognition: Converts images of mathematical equations into LaTeX code.
- High Accuracy: Specifically adapted for mathematical notation, recognizing complex symbols and handling handwriting variations.
- Efficient Inference: 4-bit quantization (bnb-4bit) significantly reduces memory footprint (0.674 GB peak memory during training) and speeds up processing, making it suitable for resource-constrained environments.
- Domain Adaptation: Fine-tuned on the
unsloth/LaTeX_OCRdataset (68,686 samples) to learn precise character recognition and correct LaTeX formatting rules.
Good For
- Mathematical OCR: Ideal for applications requiring the conversion of handwritten or image-based mathematical formulas into structured LaTeX.
- Educational Tools: Can be integrated into platforms for students and researchers to digitize mathematical notes.
- Developers with Limited Resources: Its 4-bit quantization allows it to run on GPUs with limited VRAM, such as a Tesla T4 or RTX 3060, while maintaining strong performance.