metadata

title: Lightweight Embeddings
emoji: 🌍
colorFrom: green
colorTo: green
sdk: docker
app_file: app.py

🌍 LightweightEmbeddings: Multilingual, Fast, and Lightweight

LightweightEmbeddings is a high-performance framework designed for generating embeddings from text or image-text inputs across multiple languages. Engineered for efficiency and adaptability, it offers a perfect balance between speed and accuracy, making it ideal for real-time applications and resource-constrained environments.

✨ Key Features

Multilingual Support: Seamlessly process text in over 100+ languages for truly global applications.
Text and Image Embeddings: Generate embeddings from text or image-text pairs using state-of-the-art models.
Optimized for Speed: Built with lightweight transformer models and efficient backends to ensure rapid inference, even on low-resource systems.
Flexibility: Supports multiple transformer models for diverse use cases:
- Text models: multilingual-e5-small, paraphrase-multilingual-MiniLM-L12-v2, bge-m3
- Image model: google/siglip-base-patch16-256-multilingual
Dockerized: Deploy anywhere with ease using Docker, making it production-ready out of the box.
Interactive API: Comes with a Gradio-powered playground and detailed REST API documentation.

🚀 Use Cases

Search and Ranking: Generate embeddings for advanced similarity-based ranking in search engines.
Recommendation Systems: Use embeddings for personalized recommendations based on user input or preferences.
Multimodal Applications: Combine text and image embeddings to power tasks like product catalog indexing, content moderation, or multimodal retrieval.
Language Understanding: Enable semantic text analysis, summarization, or classification in multiple languages.

🛠️ Getting Started

1. Clone the Repository

git clone https://github.com/lh0x00/lightweight-embeddings.git
cd lightweight-embeddings

2. Build and Run with Docker

Make sure Docker is installed and running on your machine.

docker build -t lightweight-embeddings .
docker run -p 7860:7860 lightweight-embeddings

The API will now be accessible at http://localhost:7860.

📖 API Overview

Endpoints

/v1/embeddings: Generate text or image embeddings using the model of your choice.
/v1/rank: Rank candidate inputs based on similarity to a query.

Interactive Docs

Visit the Swagger UI for detailed, interactive documentation.
Explore additional resources with ReDoc.

🔬 Playground

Embeddings Playground

Test text and image embedding generation in the browser with a user-friendly Gradio interface.
Simply visit http://localhost:7860 after starting the server to access the playground.

🌐 Resources

Documentation: Explore full documentation
Hugging Face Space: Try the live demo
GitHub Repository: View source code

💡 Why LightweightEmbeddings?

Performance-Oriented: Delivers rapid results without compromising on quality, ideal for real-world deployment.
Highly Adaptable: Works in diverse environments, from cloud clusters to local devices.
Developer-Friendly: Intuitive API design with robust documentation and an integrated playground for experimentation.

👥 Contributors

lamhieu – Creator and Maintainer (GitHub)

Contributions are welcome! Check out the contribution guidelines.

📜 License

This project is licensed under the MIT License. See the LICENSE file for details.