You need to stop whatever you're doing because the entire AI industry just got flipped over. So a few days ago, OpenAI, the company that everyone loves to call closed AI, the one constantly mocked for its profit pivot, did the most open thing imaginable. They dropped something that isn't just a product launch.
It's a declaration of independence for every builder out there. They call it GPTO OSS. This is a state-of-the-art AI model which is the result of billions in research with performance that goes toe-to-toe with their own flagship models.
And the craziest part, you can run it on your computer on your own terms locally. So now we know what you're thinking. Another week or another model, we're all getting model fatigue.
Whether it's Quen, Llama, Gemma, Claude, it feels like every week there's a new GPT killer launch. But this one is actually different. So this is a philosophical shift.
It's about privacy. It's about control. But this incredible news creates an immediate massive question because a model like this is just a file, a very smart file, sure, but it's still just a file.
So where do you actually get it? How do you test it, manage it, and plug it into your own projects without an army of engineers? So if you look very closely at their announcement, the very first call to action, the first link they want you to click, it doesn't point to their website.
It points to this thing right here called hugging face. So and that's the mission for today's video. We're going to dive into the single most important platform in this new open AI reality hugging face.
Forget think about it as just a place to download models. It's a GitHub for AI. So stick around because we're going to show you how you can go from hearing about a new open source model drop to running it on your own machine in minutes.
We'll break down what hugging face is, what all these buttons and tabs actually mean, what you can do with it, and finally how to download and run a model locally right here on this machine. So let's get started. All right, so let's dive in.
The first thing that we're going to do is head over to huggingface. co. And if you haven't signed up, then you need to sign up like you would for any other platform.
Now, when you first land here, it can look a little bit intimidating. All right, there's a lot of buttons, lots of tabs, but it's actually organized brilliantly. So, let's break it down tab by tab and get you up to speed.
So, first and foremost is the heart of this platform, and that is the models. All right. So, as of today, there are close to 2 million models over here.
That is insane. Now, these aren't just a random collection of files. Okay.
These are powerful pre-trained models for everything you can imagine. Whether it's text generation, image creation, audio processing, video generation, there's a model for everything. Now, obviously, you're not expected to scroll through all 2 million of them.
That's what these filters on the left offer. Now you can narrow down your search to find the perfect model based on the specific task you need, the size of the model, the framework it's built with, and even the tools that you want to use to run it. So whether that's locally on your machine or on cloud provider like Grock or Together AI, it makes finding the right tool extremely easy.
All right, so we're done with the models. Now we're going to be moving on to the data sets. Now these models are absolutely useless without data.
Okay, now think of this tab as a massive library of the raw materials used to train and test these AI models. Okay, you'll find everything here. Text data sets from books and articles, image data sets, from photos and medical scans, and also audio and video data sets.
Insane. And just like with the models, you get a bunch of filters to find exactly what you need based on modality, size, and file type. Okay, so we finished with models.
Now we finished with data sets. Now we're moving on to spaces. This is one of the team's favorite parts.
It's a place where the community can host and share live demos of their models. It's not just a code repository. It's a gallery of working interactive AI applications.
It's the best place to see what's possible right now. So, just take a look at this. So, right in your browser, you can generate images, create 3D art, and even synthesize videos.
Insane, bro. This is amazing. And of course, we have docs, we have learn, we have blogs and papers as well.
This is a huge knowledge base. Hugface isn't just giving you the tools, they're teaching you how to use them at the same time. Their documentation is topnotch and the learn section has free high quality courses on LLM, agents, diffusion models and so much more.
All right, so that's the highle talk. Models are the tools, data sets are the raw materials and spaces are the finished products. All right, so now that's enough talk for now.
Let's build something. Now the first step in any project is choosing your tool. All right, so let's head back to the models tab.
Okay, and over here you can immediately see what's trending. Insane. And it's no surprise right at the top you have to see OpenAI's models that we just talked about.
So you have this GPOSS 12B and the smaller 20B version with a new Quen image model that's not that far behind. >> For this demo though we need something that'll run smoothly on this machine. So let's use the filters.
Okay, we'll set text generation as the task. Want a model that isn't too big. So we'll filter for something under 12 billion parameters.
And importantly we'll select the GGUF library. Now, that's a specific format that makes it super easy to run with our other tool of choice, which is Olama. All right.
So, as soon as we do that, you can see that the list narrows down to some familiar names like Google's Gemma, Quen, and a bunch of others. So, we've used Google a bunch of times on this channel. So, let's try something different.
Let's go with the Quen model. Okay. Now, when you click on a model, you land on this model card.
This is important. Okay. It's the readme file for the model.
It tells you what it does, how it was trained, its limitations, and most importantly, how to use it. Never skip reading the model card because it's an exact blueprint for how you should use the specific model. What we're going to use is a fantastic tool called Olama.
If you don't have it, just head to their site and install it. Once it's set up, running a model from HuggingFace is just a simple command. I've already got it installed, so I'm just going to pull it up from my terminal.
Now, watch how simple this is. Okay, the command is run followed by the model ID from Hugging Face, which you can just copy from the top of the page. Now, Ola is going to check if I have the model downloaded.
It sees that I don't. So, now it starts pulling up the manifest and downloading the necessary files. All right.
Fast motion. Fast forward motion. And we're done.
Just like that, bro. It was that simple. >> God damn.
All right. So, just like that, we're done. And the Quen model is now running on my laptop.
Right. So, let's try something more practical. Give me a Python code snippet to connect to Postgra SQL database using site copy G2 library.
Now this what I just put in is a real world task. Okay. And perfect.
Gives me the code complete with placeholders for my credentials. Even error handling. Insane.
This is an incredibly powerful developer assistant running entirely on my own hardware. I'm not even sending any data to a third party API. It's all private, all local.
All right, so that's it, ladies and gentlemen. We've gone from being overwhelmed by the idea of an AI to taming a state-of-the-art model in just a few minutes. We saw how Hugging Face acts as a central hub.
We saw how you can test models directly in the browser. And then we downloaded and ran our own powerful model locally using Olama. This workflow fundamentally changes the game for individual creators and small teams.
So what's next? With this new leap in AI, you could build a custom chatbot for your website. You could create a tool that summarizes long articles.
You could build a code generation system fine-tuned for your specific code base. So, with that being said, ladies and gentlemen, comment what you want us to build next.
