gpt-oss: OpenAI releases 2 open-source models

https://openai.com/index/introducing-gpt-oss/

23 points · 9 comments · view on lemmy.world

9 Comments

jeena@piefed.jeena.net · 14 pts · 1y (4 replies)

I tried the 20b on my PC with ollama and Open WebUI and I have to say for some of my use cases it preforms similar to the online version, I was impressed.

glimse@lemmy.world · 2 pts · 1y (3 replies)

What card are you running it on?

jeena@piefed.jeena.net · 2 pts · 1y (2 replies)

Nvidia rtx 3060 the 12GB Vram one

glimse@lemmy.world · 2 pts · 1y (1 reply)

Not bad. Does it use all the VRAM?

jeena@piefed.jeena.net · 2 pts · 1y

I need to check with that model but usually yes.

thoon@feddit.nl · 11 pts · 1y

Finally, their company name is starting to make SOME sense

nymnympseudonym@lemmy.world · 7 pts · 1y

It's awesome but that 128k context window is a throwback to Llama3 days

I bet the closed $ource model has like 2MB context

fubarx@lemmy.world · 2 pts · 1y

Ran the 20B on a Mac under LMStudio. Pretty zippy and did OK on basic coding tasks.

Mwa@thelemmy.club · 1 pts · 1y

Crazy but sure ig