---
language: [or]
license: apache-2.0
base_model: Qwen/Qwen2.5-VL-3B-Instruct
datasets: [shantipriya/odia-ocr-merged]
tags: [odia, ocr, vision-language, lora, peft]
---

# Odia OCR — Qwen2.5-VL-3B Fine-tuned (v2)

LoRA fine-tune of **Qwen/Qwen2.5-VL-3B-Instruct** on 66K Odia OCR image-text pairs (checkpoint-3800).

**Metrics at step 3600**: CER 9.60% | Exact Match 68.2% (500 test samples)

**LoRA config**: r=128, alpha=256, 7 target modules (q/k/v/o/gate/up/down proj)

## Usage
```python
from transformers import AutoProcessor, Qwen2_5_VLForConditionalGeneration
from peft import PeftModel
import torch

processor = AutoProcessor.from_pretrained("Qwen/Qwen2.5-VL-3B-Instruct", trust_remote_code=True)
model = Qwen2_5_VLForConditionalGeneration.from_pretrained("Qwen/Qwen2.5-VL-3B-Instruct", torch_dtype=torch.float16, device_map="auto")
model = PeftModel.from_pretrained(model, "shantipriya/odia-ocr-qwen-finetuned_v2")
model.eval()
```