README.md

---
license: mit
tags:
  - peft
  - lora
  - qwen2
  - multilingual
  - ocr
  - translation
  - safetensors
base_model: Qwen/Qwen2-VL-7B-Instruct
---

# Babel

## Model Description
Babel is a Qwen2-VL LoRA adapter fine-tuned for multilingual OCR (Optical Character Recognition) and translation tasks. It can extract text from images across multiple languages and translate between them, making it ideal for document digitization, cross-language content processing, and international business automation.

## Model Architecture
- **Base Model**: `Qwen/Qwen2-VL-7B-Instruct`
- **Fine-tuning Method**: LoRA (Low-Rank Adaptation) via PEFT
- **Checkpoint**: Final checkpoint
- **Task**: Multilingual OCR + Translation (Vision-Language)

## Training Details
- **Framework**: HuggingFace PEFT + Transformers
- **Dataset**: Multilingual document images with text annotations and translations
- **Languages**: Multiple languages supported including English, Hindi, and more
- **Approach**: Vision-language fine-tuning with OCR and translation objectives

## Files
| File | Description |
|------|-------------|
| `adapter_model.safetensors` | LoRA adapter weights |
| `adapter_config.json` | PEFT adapter configuration |
| `tokenizer.json` | Tokenizer vocabulary |
| `tokenizer_config.json` | Tokenizer configuration |

## Usage

```python
from transformers import AutoProcessor, Qwen2VLForConditionalGeneration
from peft import PeftModel
from PIL import Image
from huggingface_hub import snapshot_download

# Download adapter
adapter_dir = snapshot_download(repo_id='devanshty/Babel')

# Load base model
base_model = Qwen2VLForConditionalGeneration.from_pretrained(
    "Qwen/Qwen2-VL-7B-Instruct",
    torch_dtype="auto",
    device_map="auto"
)
processor = AutoProcessor.from_pretrained(adapter_dir)

# Load LoRA adapter
model = PeftModel.from_pretrained(base_model, adapter_dir)
model.eval()

# OCR + Translate
image = Image.open("document.jpg")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "image": image},
            {"type": "text", "text": "Extract all text from this image and translate it to English."}
        ]
    }
]
text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = processor(text=[text], images=[image], return_tensors="pt").to(model.device)
output = model.generate(**inputs, max_new_tokens=1024)
print(processor.decode(output[0], skip_special_tokens=True))
```

## Download & Use

```python
from huggingface_hub import hf_hub_download
adapter = hf_hub_download(repo_id='devanshty/Babel', filename='adapter_model.safetensors')
```