Zero Shot Text Classification

This is the source code for my Huggingface Space deployment. You can checkout the deployed application on Huggingface Spaces. The task of Zero Shot Classification is a Natutal Language Processing technique where a model can classify text into categories it has not seen during training. This is acheived by training a model like DeBERTa on a variety of NLP tasks allowing the model to generalize and predict the most relevant class for a given text input based on the context provided by candidate labels.

I have used a DeBERTa model architecture that excels at Zero Shot Tasks very well. I have quantized a finetuned model, sileod/deberta-v3-base-tasksource-nli provided by Damien Sileo to further improve the inference latency. The model has been quantized using Huggingface Optimum for ONNX. You can see the quantize.ipynb for the source code for quantization.

Gradio WebUI

You can access the Gradio App directly on Huggingface Spaces by following this Link. You can embed this application as an iframe using the following HTML code:

<iframe
	src="https://arnabdhar-zero-shot-classification-deberta-quantized.hf.space"
	frameborder="0"
	width="850"
	height="450"
></iframe>

The Direct URL is: https://arnabdhar-zero-shot-classification-deberta-quantized.hf.space

API endpoint

1. cURL command

curl -X 'POST' \
  'https://arnabdhar-zero-shot-classification-deberta-quantized.hf.space/predict' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -d '{
  "text": "I am loving it",
  "labels": ["positive", "negative"]
}'

2. Postman

Base URL: https://arnabdhar-zero-shot-classification-deberta-quantized.hf.space

Method: POST

URL: /predict

Body:

{
  "text": "I am loving it",
  "labels": ["positive", "negative"]
}

3. Python

import requests
import json

# Define the URL
url = 'https://arnabdhar-zero-shot-classification-deberta-quantized.hf.space/predict'

# Define the headers
headers = {
    'accept': 'application/json',
    'Content-Type': 'application/json'
}

# Define the payload
data = {
    "text": "I am loving it",
    "labels": ["positive", "negative"]
}

# Make the POST request
response = requests.post(url, headers=headers, data=json.dumps(data))

# Print the response
print(response.json())

4. Javascript

// Define the URL
const url = 'https://arnabdhar-zero-shot-classification-deberta-quantized.hf.space/predict';

// Define the headers
const headers = {
  'accept': 'application/json',
  'Content-Type': 'application/json'
};

// Define the payload
const data = {
  text: 'I am loving it',
  labels: ['positive', 'negative']
};

// Make the POST request
fetch(url, {
  method: 'POST',
  headers: headers,
  body: JSON.stringify(data)
})
.then(response => response.json())
.then(data => console.log(data))
.catch(error => console.error('Error:', error));

Running the Application

To run the application, I highly recommend using Docker but there are also other ways to get started. But first you have to clone the repository onto your local machine.

$ git clone https://github.com/arnabd64/Zero-Shot-Text-Classification.git
$ cd Zero-Shot-Text-Classification

Using Docker

1. Build the Docker Image:

$ docker build -t zero-shot-text-classification:latest .

2. Run the following docker command:

$ docker run -itd \
-p 8000:8000  \
-e HF_MODEL=pitangent-ds/deberta-v3-nli-onnx-quantized \
-e PORT=8000 \
-e WORKERS=2 \
zero-shot-text-classification:latest

After running the docker container wait for a few minutes for the model to download and load into the memory.

Using Docker Compose (Recommended)

$ docker compose up -d --build

After running the docker compose wait for a few minutes for the model to download and load into the memory.

Using python

Note: Use python 3.11 or later (Python 3.10 should work but I have not yet tested it)

1. Create a virtual environment and install dependencies

# create a virtual environment
$ python -m venv .venv

# activate the environment
$ .venv/bin/activate # on linux and macOS
$ .venv/Scripts/Activate.ps1 # on Windows

# install dependencies
$ pip install -e .

2. Set the following Environment Variables

HF_MODEL=pitangent-ds/deberta-v3-nli-onnx-quantized
PORT=8000
WORKERS=2

3. Run the application

$ python main.py