Cohere tutorial: How to create a dog's breed recognition API

Friday, December 02, 2022 by jakub.misilo

Why build an AI app using Cohere API?

In today's prodigious sea of information, discerning an animal or its breed from a mere description can be a Herculean task. Embrace the magic of Cohere AI—a groundbreaking tool poised to revolutionize data interactions.

Embark on a fascinating CoHERE tutorial quest to forge an API that pinpoints a dog's breed, fueled by a brief description. Powered by Cohere API and with visuals generated by DALL·E 2, this venture showcases the true potential of AI in crafting a Cohere app.

Eager to delve deeper into the captivating world of Cohere AI? Our treasure chest of Cohere tutorials awaits! And for those yearning for communal learning, our remarkable AI Hackathons unite global enthusiasts to experiment with this and other tutorials. Seize the opportunity to learn, collaborate, and innovate like never before!

🏹 Let's start

We need to start by creating a directory for our project. We will call it dog-breed-recognition and then we will create a virtual environment for it.

mkdir dog-breed-recognition
cd dog-breed-recognition
python3 -m venv venv
.\venv\Scripts\activate

We have to create Cohere and OpenAI account, and then download API Keys. We will use them for APIs authorization. Let's create .env file and put them there:

COHERE_API_KEY=&lt;COHERE API KEY&gt;
OPENAI_API_KEY=&lt;OPENAI API KEY&gt;

Now let's install all the necessary libraries:

pip install python-dotenv
pip install --upgrade cohere
pip install openai
pip install fastapi[all]

Now we can create an app.py file and start writing our code.

🤖 Coding!

First, we need to import all the necessary libraries and load environment variables:

import os
import cohere
import openai
from dotenv import load_dotenv
from fastapi import FastAPI
 
# load env. variables
load_dotenv()

Then we need to create a FastAPI app and authorize Cohere and OpenAI clients:

app = FastAPI()
 
# initialize the Cohere Client with an API Key
co = cohere.Client(os.getenv('COHERE_API_KEY'))
openai.api_key = os.getenv('OPENAI_API_KEY')

Now I will define a prompt for Cohere's LLM. It will be used to generate a prediction of a dog's breed. I will pass some examples of dog's breeds and their descriptions to the model. I will prepare space for our description, which will be passed to the model.

def generate_prompt(description):
    prompt = f'''Please predict the dog's breed in the description.
 
Description: This dog is a medium-sized breed of dog that was developed in the United States. The breed is often used as a working dog on farms and ranches, and is also a popular companion animal.
Breed: Australian Shepher
--
Description: This dog is a cheerful, high-spirited breed that is well suited for a variety of activities, including hunting, agility, and obedience. They are very intelligent and easily trained, and are excellent companions for active families. They are also known for their friendly dispositions and love of children.
Breed: Brittany
--
Description: They are large, muscular dogs with a short, thick coat. They have a wide head and a short, square muzzle. Their eyes are small and their ears are floppy. They have a thick neck and a strong, muscular body. Their tail is short and their legs are muscular.
Breed: American Bulldog
--
Description: This dog is a medium-sized breed of dog that is native to Poland. They are intelligent and active dogs that are known for their herding abilities. They have a thick, shaggy coat that can be either black, white, or brindle in color. They are loyal and protective dogs that make great family pets.
Breed: Polish Lowland Sheepdog
--
Description: They are large, working breed of dog known for its obedience, loyalty, and intelligence. They are strong, agile dogs with a well-proportioned build. They have a long head, erect ears, and a long, thick coat. The breed is intelligent and easily trained, making them excellent working dogs. They are used in a variety of roles, including law enforcement, search and rescue, and as assistance dogs.
Breed: German Shepherd Dog
--
Description: They are a medium-sized breed of dog, originally bred in England as gun dogs. They are bred to be lovable companions and are often considered one of the best breeds for families. They are relatively easy to train and are known for being intelligent and eager to please. Cockers have a reputation for being one of the most affectionate dog breeds, and are also known for being good with children.
Breed: Cocker Spaniel
--
Description: {description}
Breed:
'''
    return prompt

Now we can define an endpoint for our prediction. We will use Cohere's LLM to generate a prediction. We will pass our prompt to the model and then we will return the result. Then the result will be used for generating a photo.

@app.get('/')
def breed_prediction(description: str):
    prompt = generate_prompt(description)
    response = co.generate(
        model='xlarge',
        prompt=prompt,
        max_tokens=5,
        temperature=0.75,
        stop_sequences=['--'],
    )
 
    breed = response.generations[0].text.strip()
 
    img = openai.Image.create(
        prompt=f'Image of {breed}',
        n=1,
        size='512x512'
    )
 
    img = img['data'][0]['url']
 
 
    return {
        'breed': breed,
        'img': img
    }

🚀 Running the app

Now we can run our app:

uvicorn app:app --reload

We can test our app by sending a request to the endpoint. I will use Postman for that. It's a great tool for testing APIs. We can download it here. That's how our request URL should look like:

localhost:8000/?description=&lt;description&gt;

My description looks like:

This is a large, white, fluffy dog with a thick coat. They are very friendly and love being around people. They were originally bred in Siberia to help people with their daily chores, such as pulling sleds and herding reindeer.

🧑‍🚀 Let's check results!

Tutorial accompaniment image — Prediction: 'Siberian Husky'

The results speak for themselves! It's truly remarkable how much insight we can extract from text today. We encourage you to try your own descriptions and share your results on our lablab.ai Discord!

In this Cohere tutorial, we've shown you how to build a Cohere app, specifically a dog breed recognition API. But the journey doesn't stop here. You can join lablab.ai's AI hackathons to put your new skills to the test in a live setting. It's a fantastic opportunity to learn, share, and innovate alongside a community of like-minded individuals from around the world.

And remember, knowledge is power. Progressing your knowledge in this rapidly growing industry could be the catalyst for a career change. So why wait? Join the AI revolution with lablab.ai and start building with the Cohere API today!