00:00Can you run DeepSeek R1
00:02On your own machine?
00:02I am also gonna try this out on the RTX 5090
00:05So first let's do the 1.5 billion parameter model
00:08This is the prompt I'm gonna give it
00:09A farmer has 17 sheep all but 9 die
00:12How many sheep are left?
00:13So obviously the answer is 9
00:16I get about 111 tokens per second
00:19So it's saying that there should be 8 sheep remaining
00:22Not gonna lie when I skimmed the prompt
00:23I thought the same thing
00:24Maybe I am just dumb too
00:25The thing is that it was really fast
00:27But the answer was wrong
00:28But for a 1.5 billion parameter model, honestly props.
00:32The 7 billion parameter model.
00:3490 tokens per second?
00:35It got the answer wrong.
00:37But next up we're going to do the 14 billion parameter model.
00:40So a bit slower, but it got it right and it was smart.
00:44That's all that matters.
00:45And finally, the 32 billion parameter version.
00:50And it gets it.
00:50Now 30 tokens per second seems slow but honestly for how powerful this model is, that's really impressive.
00:56Running locally is just the way to do it.