There’s so much cope around DeepSeek.

They didn’t just release an incredible model; they also introduced a groundbreaking training method (R1 Zero) that’s already being replicated.

I highly doubt they lied about the training costs, but even if they did, they’re still amazing for giving the world this incredible gift.

Their models are also open source. You can download them.

https://www.deepseekr1.org/en/download

Reply to this note

Please Login to reply.

Discussion

I tried it out, it's pretty wild.

Just don't ask it about China.

How are people running it on Android? Their app doesn't look open source.

You can use this website. Go to settings, and choose any model. It will download the model, keep it in the browser cache, and run it on your native hardware using WebGPU on any device.

They don’t have the latest smaller model yet. I’ll share the link for that later, so you can add it to this web app.

https://chat.webllm.ai/

если один блок имеет информацию о предыдущем то в будущем есть информация о прошлом.