https://github.com/jisnap/fine_tuning_pretrained_models
https://github.com/jisnap/fine_tuning_pretrained_models
autotokenizer datasets huggingface transformers
Last synced: 24 days ago
JSON representation
- Host: GitHub
- URL: https://github.com/jisnap/fine_tuning_pretrained_models
- Owner: JisnaP
- License: mit
- Created: 2025-02-27T13:56:24.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2025-02-27T18:29:52.000Z (over 1 year ago)
- Last Synced: 2025-02-27T22:11:09.101Z (over 1 year ago)
- Topics: autotokenizer, datasets, huggingface, transformers
- Language: Jupyter Notebook
- Homepage:
- Size: 35.2 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# 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**
```bash
git clone https://github.com/JisnaP/Fine_tuning_pretrained_models.git
cd Fine_tuning_pretrained_models
```
### 2️⃣ **Create a virtual environment **
```bash
python -m venv venv
source venv/bin/activate
```
### 3️⃣ **Install dependencies**
```bash
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**
```bash
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:
```json
{
"sentence1": "The weather is nice today.",
"sentence2": "It's a beautiful day outside."
}
```
### **3️⃣ Example API Response**
```json
{
"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](https://fastapi.tiangolo.com/)
- [Hugging Face Transformers](https://huggingface.co/transformers/)