---
license: apache-2.0
base_model:
  - Qwen/Qwen3-4B-Instruct-2507
---

Cantonese Fine-Tuned Qwen 4B Model

This repository contains a **fine-tuned version of Qwen 4B** optimized for **Cantonese language understanding and generation**, along with multimodal extensions.

---

## 🔗 Model Components

- **Base Model**  
  [Qwen3-4B-Instruct-2507](https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507)

- **Cantonese Fine-Tuned Model**  
  [qwen4b-instruct-cantone-ft](https://huggingface.co/Issactoto/qwen4b-instruct-cantone-ft)

- **Vision Encoder (CLIP)**  
  [clip-vit-base-patch32](https://huggingface.co/openai/clip-vit-base-patch32)

- **Projection Layer (Multimodal Alignment)**  
  [qwen4b-instruct-image-project](https://huggingface.co/Issactoto/qwen4b-instruct-image-project)

---

## 📊 Datasets

- **🇭🇰 Cantonese Instruction Dataset**  
  [yue-alpaca](https://huggingface.co/datasets/hon9kon9ize/yue-alpaca)

- **🖼️ Flickr8k (Cantonese Translated)**  
  [flickr8k-cantonese](https://huggingface.co/datasets/Issactoto/flickr8k-cantonese)

---

## 🚀 Try it on Google Colab

Easily test the model in your browser:

[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1syJD8xrvtHVUJJKlkahAS83khKJIhB7B?usp=sharing)

---