---
license: apache-2.0
language:
- vi
- en
tags:
- crochet
- vision
- multimodal
- patterns
- translation
- qwen2-vl
---

<p align="center">
  <img src="logo.png" width="300" height="300" alt="Purlie-2B Logo">
</p>

# ✨ Purlie-2B: The Crochet Translation Model ✨

**Purlie-2B** is a specialized multimodal vision-language model fine-tuned for the crochet community. It excels at translating Vietnamese crochet patterns into standard English terminology and can extract pattern instructions directly from images.

Whether you're looking at a handwritten chart, a social media screenshot, or a text-based pattern, Purlie-2B understands the unique shorthand and technical nuances of crochet.

---

## 🚀 Key Features

- **🇻🇳 Vietnamese to 🇺🇸 English Translation**: Translates technical crochet terms (e.g., "mũi móc đơn" to "single crochet") and abbreviations.
- **🖼️ Image-to-Pattern Extraction**: Recognizes crochet patterns within images and converts them to structured text.
- **🧶 Technical Expertise**: Understands specific techniques like BLO (Back Loop Only), FLO (Front Loop Only), Magic Rings (MR), and complex stitch combinations.
- **⚡ Optimized for Inference**: Compatible with Unsloth and standard Hugging Face transformers.

---

## 📊 Example Usage

### Text Translation (Vietnamese to English)

**Prompt:**
> Dịch đoạn mã móc len sau sang tiếng Anh:
> Hàng 1: MR, 6 mũi móc đơn (6)
> Hàng 2: 6 mũi tăng (12)
> Hàng 3: (1 mũi móc đơn, 1 mũi tăng) x 6 (18)

**Purlie-2B Response:**
> Row 1: MR, 6 single crochet (6)
> Row 2: 6 increase (12)
> Row 3: (1 single crochet, 1 increase) x 6 (18)

---

### Image Extraction & Translation

**Prompt:**
> Dịch những gì bạn thấy trong hình ảnh này sang tiếng Anh chuyên ngành móc len.

**Purlie-2B Response (based on pattern_test_image.jpg):**
> Crochet pattern for a cute cat face charm
>
> 1. White yarn
> 1. Mr 6x
> 2. 6v
> 3. (x,v)6
> 4. (x,v,x)6
> 5. 4x, 3TV, x, FE, FT, 4sl, TF,EF, x, 3TV, 4x
> 6. (2ch,sl) in various colors
>
> Tiktok: Bích Ngọc

---

## 🛠️ How to Use

You can load Purlie-2B using the `unsloth` library for fast inference or standard `transformers`.

```python
from unsloth import FastVisionModel

model, tokenizer = FastVisionModel.from_pretrained(
    model_name = "PinkPixel/Purlie-2B",
    load_in_4bit = False,
)
FastVisionModel.for_inference(model)
```

---

## 📜 License

This model is licensed under the **Apache 2.0 License**. See the `LICENSE` file for details.

---

## 💖 Credits & Support

**Made with ❤️ by Pink Pixel**

- **Website:** [pinkpixel.dev](https://pinkpixel.dev)
- **GitHub:** [github.com/pinkpixel-dev](https://github.com/pinkpixel-dev)
- **Buy me a coffee:** [Coffee Link](https://www.buymeacoffee.com/pinkpixel)

*Dream it, Pixel it* ✨