So, OpenAI recently launched their open source models. So, in this video, we will talk about that recently, GPT, that is the models owned by OpenAI, they have recently opened some of their models. So, in this video, we will see what it is exactly. And it is a very revolutionary thing.
You know the GPT, that is that powers your charged GPT. And in AI, there is a dominating transformer engine. And in AI, there is a dominating transformer engine.
And in AI, there is a dominating transformer engine. And in AI, there is a dominating transformer engine. And in AI, there is a dominating transformer engine.
And in AI, there is a dominating transformer engine. And in AI, there is a dominating transformer engine. And in AI, there is a dominating transformer engine.
And in AI, there is a dominating transformer engine. an open source version is available. So with that, let's start with the video.
So today, that is August 5, 2025, Aaj Ke Din, GPT says introducing GPT-OSS. So what is in this? That there are two models which are now available for open source.
That is GPT-OSS 120 billion parameter and GPT-OSS 20 billion push the frontier to open way reasoning models. So they have a hugging face link available, I will show you that later. So this is they are available on hugging face. You can see this is the model.
And let's go ahead and see what it is exactly. Okay So we are releasing GPT-OSS 120B and this model to state of the art open weight language models that deliver strong real world performance at low cost. So they have launched these two models completely open source and it is under Apache license. That means that companies can use it, see its code and even modify it and deploy it. That's a huge thing.
Okay, let's move on. What else do we have? Available under the flexible Apache 2 license, the model outperforms similarly size compared to open models on reasoning tasks, demonstrate strong tool capabilities and are optimized for efficient deployment on consumers' hardware.
This is a very revolutionary change in the AI world. If we scroll down a little further, then here we have some diagrams. I mean, you can go through these things. So this is the...
comparison and evaluations so if we look at it from the perspective of coding that how these models perform so you can see that this is one this is the OSS model this is also OSS model, this is also an OSS model with tools, without tools and this is their comparison with O3 O4 mini which are basically reasoning models. So O3 are their reasoning models and O4 they had recently launched which are very high capable. If we look at comparison here, that is this one, that is GPT OSS with tools, it almost almost performs as much as O3 and O4 Mini do. So that means, if you have ever used GPT-4 or O Mini, which is really fast and really nice, then it almost achieves that much accuracy, which is really good benchmarks.
So these benchmarks are really good benchmarks. And if we look ahead, you can see here there are some more benchmarks. So Humanity last exam, so here I can see the performances down. But if we look further, their performance is even better in many places. So here is a tweet by Sam Altman.
So this is the tweet that GPT-OSS is a big deal. That is true. And it's a state-of-the-art open weights reasoning model with strong real-world performance compared to O4 Mini. So GPT-OSS is a reasoning model. So an open source reasoning model.
So last time when an open source reasoning model that was like deep-seek, it was a huge game-changing thing. and now GPT which is OG in this particular field we are getting an open source model from them. That's actually a big thing. You can run locally on your own computer or phone with a smaller size. We believe that this is the best and most usable open model in the world.
So, you can read this tweet here. The open AI's mission to ensure AGI that benefits all humanity. To that end, we are excited for the world to be building an open AI stack created in the United States based on democratic values available free to all and for wide benefit.
Now, what will be the main changes because now it is open-weight. The thing is that its adoption rate is going to be very high. The companies are going to use it crazily. The people are going to use it crazily.
There will be more models in which people are fine-tuning these models. So, that was the main news that now it is open-source and I really, really want to use this particular model. But now what I can see is that do 120 billion parameter. It's actually a production general purpose high reasoning model and for this it will take a good GPU. So for companies for production use it's a Really nice thing.
And for low latency, local and specialized like 21 billion parameters with 3.6 active billion parameters. So, this is for the lower latency. It's like, you know, a mini model.
Means, if you want to run locally or if you want to train for a specific thing, then that is the thing. Okay. So, you can see what all is available. Okay. We also have inference examples here.
How to use this? So, you have to install transformers. That is pip install transformers. Right. You have to make a pipeline for text generation.
You have to give model ID here, that is this model ID. And then it's same. As we call openAI, it's GPTML format. In which role user content is there and then we have to pipe the output. This is called chat ML format.
So, there are prompting types. This is chat ML format. And after that, if we look ahead, we can use transformer serve.
So, documentation is pretty nice. And all the code is there. Both GPT OSS models can be fine-tuned for a variety of special things.
So now you are going to see a bombard. You will get a whole flow. Many companies will bring their fine-tuned model on top of this. So it was a huge step from OpenAI team. Let's see how the world reacts on it.
But it is going to be a game-changing thing. So stay updated with these kind of news and let's see how AI world transforms. And by the way, just in case you are curious about learning about AI and stuff. So we have recently launched a Gen AI cohort in JavaScript, whose links you will find in the description.
So there we talk about all these things, how these LLMs work, how to orchestrate these LLMs, how to build custom tools for these LLMs in Gen AI. So in that, we talk about agentic workflows. So just in case you are very curious about the AI and you want to learn about AI, so all the links are in the description as we have our course already up and running.
So do check out the links in the description. in the description. So with that, let's end the video.
How was the video? Do let me know in the comments. I'll see you in the next video.
Until then, bye-bye and take care.