OpenAI's nightmare: DeepSeek R1 on a Raspberry Pi

Blog post with links to how I ran it with ollama on Pi 5's CPU (Qwen Distilled 14b) and an AmpereOne (671b), and llama.cpp + Vulkan on AMD GPU:

And direct link to the DeepSeek benchmark results (yes I did run on both the Pi CPU alone and on eGPU):

OpenAI is betting on a little-known startup to stay ahead of Elon Musk in the supercomputer race.

Might be sensitive question, if you run it local, will you still worry about the "called" security issue without use their cloud?

No, not when running it through Ollama or llama.cpp at least.

@nvidia Could you please send him an H100? I would really like to see that GPU on top of an Raspberry PI 5 🔥

It's a cool enough feat without the clickbait title man.

> No confusion about who precisely is Winnie the Pooh Jeff’s videos are now banned from behind the Great Firewall

How Qwen relates to DeepSeek? It's not clear to me

The Qwen-Distilled-14b model is distilled using DeepSeek to make it a more useful model than standard Qwen 2.5. And Ollama labels distilled models as "deepseek-r1" still lol

Can you please make a video on increasing my social credit score using my raspberry pi?