So, OpenAI recently released the Codex toolkit, which enables developers to train and deploy AI models without having to write any code. As of today, you can get access to Codex by joining their waitlist here. It took me about a week to get a response.

Hugging Face is a company which aims for the decentralization of AI. They develop tools for those building applications using machine learning. It’s most notable for its Transformers library built for natural language processing applications and its platform that allows users to share machine learning models and datasets.

Anyway, I’m going to show you how to use Codex and the Hugging Face library to train a text summarization model in just 5 minutes!

Let’s get right into it

Assuming you have made a Hugging Face account and access to Codex, you have to set up a “Space” in Hugging Face. Spaces on Hugging Face allow you to host ML demos on your profile and it supports 3 frameworks that can quickly build Python apps: Streamlit, Gradio and Static.

1. Create a new Space.

2. Name it however you want and then select Gradio as your Space’s SDK (Software Development Kit). Ignore the licenses for now and public or private is up to you.

3. Click on “Files and versions” and add a new file.

4. We are going to name this file “requirements.txt” and type “torch” in one line and “transformers” in the other. We need access to these two libraries for our application to work. Scroll down and click “Commit new file”.

5. Good job! Now we are going to head over to Codex and generate the python code we need to run our summarizer model.

Please keep in mind Codex is still in beta and should not be heavily relied on for generating proper code. It sure can help you get started however, and soon in this tutorial I’ll show you why you need some level of coding knowledge before heavily relying on tools such as Codex.

So in the OpenAi Playground , after your account has been given access to Codex, you’ll need to switch the model to “code-davinci-002”

Then you’ll input what I have highlighted below as a prompt to generate the code. Use #’s when giving instructions to Codex so it knows what you are typing is not code itself.

Cool right!

Yea! But if you were to run this code alone you’d get an error saying “Error occurred while trying to proxy: huggingface.co/”

6. You’re going to have to add “inline = False” as seen below before copying the code.

This is because your typical Gradio code is something you would usually write on Google Collab and then launch the UI (user interface) in your collab notebook itself. However, that’s not what we are doing today as we are running our model on Hugging Face so that we can have it on display in our Space for anyone to visit and demo.

7. Copy the code from the Playground and head back over to Hugging Face. Name your whatever you want but make sure to add “.py” at the end of it so Hugging Face knows to run this file as Python app. Scroll down and then click “Commit new file”.

8. Click “App” to preview your demo and wait till the “Building” progress bar says it’s “Running”.

9. Test it out! Put any body of text you’d like summarized and see if your application works!

10. Good job!

11. Proud of you.

Now you’ve made your first ML demo that you can share with anyone test out simply by sending them a link to your space.

If you want more posts similar to this, visit CloudCap Solutions and join the signup for my newsletter. It hasn’t launched as yet but I plan on only sending articles related to AI and Web3 topics including articles and recent developments along with my personal opinions on those topics.

I hope this inspires you to explore Hugging Face on your own as well after showing you how easy it was to build, deploy and host a ML Demo app on their website under your own profile.

Thanks for reading! If you found this useful feel free to visit CloudCap Solutions, share this with others, or just leave a clap or comment! All feedback is warmly welcomed.

See ya!