README.md

metadata

base_model:
  - Qwen/Qwen2.5-VL-3B-Instruct

🧾 Qwen2.5-VL Invoice OCR - LoRA Adapter

Fine-tuned LoRA adapter for Qwen2.5-VL-3B-Instruct specialized in extracting structured JSON data from invoice and receipt images.

Performance

• Training Loss: 35.78 → 0.024 (99.3% improvement)
• Eval Loss: 0.010
• Steps: 150+ with engineered prompt
• Dataset: 2,043 samples

Usage

from transformers import AutoModelForVision2Seq, AutoProcessor
from peft import PeftModel

base_model = AutoModelForVision2Seq.from_pretrained("Qwen/Qwen2.5-VL-3B-Instruct")
model = PeftModel.from_pretrained(base_model, "kahua-ml/invoice1")