🥘 Llama-3.2-11B-Vision-Food-mini

A highly optimized, 16-bit merged Vision-Language Model (VLM) fine-tuned for Indian Food Recognition and Analysis.

This model is a fine-tuned, 16-bit merged version of unsloth/Llama-3.2-11B-Vision-Instruct. It was trained using the Unsloth library for rapid and memory-efficient QLoRA fine-tuning. The model specializes in multimodal tasks related to Indian food, such as visual classification, detailed description, and Visual Question Answering (VQA).

---

🚀 Model Details

Property	Value
Model Type	Vision-Language Model (VLM)
Base Model	Meta Llama-3.2-11B-Vision-Instruct
Fine-Tuning Method	Unsloth QLoRA (Merged to 16-bit)
Parameters	11 Billion
Quantization	Full 16-bit (merged from 4-bit LoRA)
Owner	Anachronox39
Repo ID	Anachronox39/Llama-3.2-11B-Vision-Food-mini

Intended Use

This model is intended for research, academic and non-commercial applications focused on:

• Indian Food Classification/Recognition from images.
• Multimodal Food Description (e.g., generating detailed captions based on an image).
• Visual Question Answering (VQA) on Indian food content (e.g., "What region is this dish from?").
• Note: Commercial use is strictly governed by the Llama 3.2 Community License.

Training Data

• Dataset Used: SKSKCulinaryAI/Indian_Food_Images_Dataset
• Training Hardware: 2X Dual NVIDIA GeForce RTX 3060 GPUs

---

💻 How to Use for Inference

The model is loaded using the FastVisionModel from the unsloth library for optimized inference performance.