This project fine-tunes a language model entirely within a single Jupyter Notebook to extract structured fields — name, category, manufacturer, and price — from product descriptions.

✅ All logic — dataset prep, model training, and inference — is inside product_info_finetune.ipynb.

Input:

Maggi noodles pack of 2 by Nestlé for $0.50

Expected Output:

{
  "name": "Maggi noodles",
  "category": "",
  "manufacturer": "Nestlé",
  "price": "$0.50"
}

✔ Prepares a noisy, realistic dataset of product descriptions
✔ Handles missing fields gracefully by returning ""
✔ Fine-tunes a model (e.g., Mistral, LLaMA, etc.) using TRL or Unsloth
✔ Runs inference to verify accuracy

• Python 🐍
• Hugging Face Transformers + Datasets
• Unsloth (for fast fine-tuning)
• 🤗 TRL / SFTTrainer
• JSONL formatted training data

from transformers import pipeline

pipe = pipeline("text2text-generation", model="your-model-name")

res = pipe("Colgate toothpaste 200g by Colgate-Palmolive for $3")
print(res[0]["generated_text"])

Expected:

{
  "name": "Colgate toothpaste",
  "category": "",
  "manufacturer": "Colgate-Palmolive",
  "price": "$3"
}

├── Fine-Tune.ipynb   # 💥 Everything is in here
├── Data.jsonl            # (optional) saved dataset
├── README.md                     # you're reading this

MIT License — free to use, modify, and share.

Built by Puneet Rawat.
Feel free to fork, star, or open issues for suggestions or improvements.