Paraphrase Detection API using FastAPI & BERT

🚀 Overview

This FastAPI application serves a BERT-based model (bert-base-uncased) fine-tuned for sentence pair classification, determining whether two sentences are paraphrases.

📌 Features

• FastAPI for serving the model
• Tokenization using Hugging Face Transformers
• Inference using PyTorch

---

🛠 Installation

1️⃣ Clone the repository

git clone https://github.com/JisnaP/Fine_tuning_pretrained_models.git
cd Fine_tuning_pretrained_models

2️⃣ **Create a virtual environment **

python -m venv venv
source venv/bin/activate  

3️⃣ Install dependencies

pip install -r requirements.txt

📌 Setup Instructions 1️⃣ Run Colab Notebook to Fine-tune and Save Model Before running the API, you must first run the Colab notebook to train and save the model parameters.

Open and run the Google Colab notebook: fine_tuning_pretrained_models_using_trainer_api_for_paraphrase_sentences.ipynb Save the trained model and download it. (You will have use these as weights in the model in app.py) Move model.pth to the models/ directory in your local repo.

🎯 Running the API

1️⃣ Start the FastAPI Server

uvicorn app:app --host 0.0.0.0 --port 8000

• The API will now be available at: http://127.0.0.1:8000
• You can test the API using the interactive Swagger UI at: http://127.0.0.1:8000/docs

2️⃣ Example API Request

Send a POST request to /predict with a JSON payload containing two sentences:

{
    "sentence1": "The weather is nice today.",
    "sentence2": "It's a beautiful day outside."
}

3️⃣ Example API Response

{
    "paraphrase": true,
    "confidence": 0.89
}

• paraphrase: true means the sentences are paraphrases
• confidence: 0.89 indicates an 89% confidence level

---

🔹 Model Used

This API uses the BERT base uncased (bert-base-uncased) model fine-tuned on a paraphrase detection dataset glue/mrpc.

If the model is not downloaded, it will be automatically fetched from Hugging Face.

---

📜 License

This project is open-source and available under the MIT License.

---

🔗 References

• FastAPI Documentation
• Hugging Face Transformers