Скачать или смотреть gpt-oss OpenAI's Open Models

OpenAI made a big move. They shared parts of their models for free, claiming they're open source. But it's a bit tricky.

You cannot run these models just anywhere, especially with 120 billion and 20 billion parameters. Forget trying to use them on a Raspberry Pi. You need a strong computer, like a high-performance desktop or a new laptop. OpenAI provides two model sizes.

The smaller one, 20 billion parameters, can run on more systems. To try it, you might want to use the Transformers pipeline from Hugging Face with PyTorch. This setup uses NVIDIA GPUs for speed and is great for data centers.

There is an easy trial option where no download is needed. Make sure to click the show reasoning button. This shows how the model thinks, which is new.

OpenAI used to keep this part secret. Now you can even see sample source code in Markdown. That's cool! To run these models, start with Olama.

It's simple: choose your OS, install, and use the commands to get the model running. OSS in GPT OSS stands for open source software, and GPT is generative pre-trained transformer. While OpenAI does not own the term GPT, they've grabbed attention in that space.

You can use a command like Olama pull GPT OSS and 20 billion to set it up on your machine. Since I do not have fancy GPUs, it runs using my CPU, but it's slow. The model takes time to think, but it shows its thought tokens.

For production, use the Transformers pipeline with PyTorch on a powerful GPU in a data center on platforms like Azure, AWS, or Google Cloud. Finally, with OpenAI's open source models, you can create cool applications.