E5 Embedder API

This is a simple FastAPI-based service that provides two main endpoints:

• /embed: Generates normalized sentence embeddings using the intfloat/e5-large model.
• /count-tokens: Counts the number of tokens for each input sentence using the Hugging Face tokenizer.

---

🔧 Requirements

• Docker installed

---

📦 Clone and Prepare the Model

To include the model in the build context, clone the Hugging Face repository inside the app folder:

git clone https://huggingface.co/intfloat/e5-large app/e5_model

Make sure the directory structure is like this:

.
├── app
│   ├── main.py
│   ├── requirements.txt
│   └── e5_model
│       ├── config.json
...

🛠 Build the Docker Image

You can build the Docker image and assign it a specific version (e.g. 0.2.0):

docker build -t e5-embedder:0.2.0 .

---

▶️ Run the Container

To run the service and expose it on port 8000:

docker run -p 8000:8000 e5-embedder:0.2.0

The API will be accessible at: http://localhost:8000

---

📬 Available Endpoints

POST /embed

Request:

{
  "sentences": ["What is the capital of France?", "Tell me about Python."]
}

Response:

{
  "vectors": [[...], [...]]
}

Each vector is a normalized embedding of the input sentence.

---

POST /count-tokens

Request:

{
  "sentences": ["Hello world!", "This is a test."]
}

Response:

{
  "token_counts": [
    {"sentence": "Hello world!", "token_count": 4},
    {"sentence": "This is a test.", "token_count": 6}
  ]
}

---

📁 Project Structure

.
├── Dockerfile
├── app
│   ├── main.py
│   ├── requirements.txt
│   └── e5_model/         # Cloned model files here

---

🧼 Clean Up

To remove the container after testing:

docker ps -a          # Find container ID
docker rm <container_id>

To remove the image:

docker rmi e5-embedder:0.2.0

---