Name: DeepSeek R1
Uploaded: 2025-01-29T20:24:47.450Z
Duration: 2 min 33 s
Description: An open-source large language model excelling in reasoning, math, and coding tasks with MIT licensing for free use and modification.

deepseek-r1

An open-source large language model excelling in reasoning, math, and coding tasks with MIT licensing for free use and modification.

DeepSeek R1

Everybody's talking about DeepSeek R1. But do you want to try it out? Let me show you exactly how. And I'm going to show you multiple ways. Here we go. 

So the first way to try it out is directly through deep seek. You go to chat.

Deepseek.com and after logging in you go to the interface it looks just like Chad GPT. Here you just select this deep think. This is the Deep seek R1 model and then it also has search. So if you wanted to be able to search the web while having the thinking model that's how to do it. 

Now here's the thing if you're like me you're going to be a little bit concerned about privacy and security given that this model is hosted on China's servers. So just assume anything that you ask it is going to be stored and analyzed by the CCP. If you don't want that let me show you a couple better options. The second option is grock grq blazingly fast inference speed and they just launched the new DeepSeek R1 model now.

Full disclosure I am a very small investor in GR. So you just go to gro.com GQ and you come right here. You select the DeepSeek R1 distill llama 70b. Now this is not the full DeepSeek R1 model. It is a distilled version but it still works incredibly well. It still thinks. It still has the thinking capability and you can run it at 250 tokens per second. 

Let me show you how fast it really is. So write the Game Tetris in Python and there it is. Look all of that thinking all of that Chain of Thought is happening in seconds. The whole thing output in seconds. It is so fast and we reached 275 tokens per second. 

So if you want blazingly fast inference on a DeepSeek distilled model, through Groq. But what if you want to run it?

Locally then you can use my favorite local inference app LM Studio. Another disclosure I am a very small investor in LM Studio as well. And by the way I'm not just trying to show the products that I've invested in. They just happen to be my favorite for each of these use cases. So go to LM Studio.

A download the one that's appropriate for you and then let me show you how to get it up and running. All right, once you download and install LM Studio, this is what it looks like. What you're going to do is come on over here to the Discover tab, click it, and then you simply search for DeepSeek R1. 

We have multiple versions of it. Now, you're not going to be able to run the full DeepSeek R1 model locally unless you have multiple beastly GPUs or you have a ton of Mac minis all wired together. Basically, very very expensive. But you can use the distilled versions locally and it'll be very fast. Here, just by searching DeepSeek, we have the R1 distill of quen 7B and we also have the R1 distill of llama 8B. 

So let's say I want to download this one I simply click on it and then on the right side it gives you different download options. So you should basically look for the least amount of quantization and to determine that you look for the highest Q number. So here you see Q8 which is less quantization than the Q4. Less quantization means usually higher quality. It probably doesn't make a difference; you probably won't notice a difference. So you could really download whatever one it suggests to download. I usually go for the Q4 km and you can see I already downloaded it. 

Now these little green icons right here signal that you will have full GPU offloading meaning it's going to put the whole model onto your GPUs vram if you have a.

Little bit of offloading. It'll be just a little bit slower. But basically look right here and it will describe how well it's going to be able to run the model on your particular machine. That's a nice thing about LM studio. 

Now there's also other versions if you do have B your gpus. So you can see there's the distill quen 14b, 32b, and then there's also the Llama 70b distill. You can download any of them as long as you can run them on your machine. 

Now once you've downloaded it, let me show you how to run it. So what you're going to do is you're going to click select a model right there and then I'm going to select DeepSeek R1 distill 8B and just click on it and then load model and it's going to load it right.

Now and then you're pretty much good to go. I'll show you. So I have an RTX 590 with 32 GB of vram, so keep that in mind as I'm showing you the inference speed that we're about to get. So there we go, it loaded. Let's start a new chat. 

So here we go, write the Game snake in Python thinking, and you can actually see the Chain of Thought right there. And this is blazing fast. This is fast by any measure. Look at that beautiful. This is a full thinking model running locally, and there we go running at 77 tokens per second. 

So now you know how to run it fully hosted with DeepSeek. You can run it on groq with insane inference speeds, and that is a US company, so you're not giving all of your data to China. And then if you

Really don't want to give your data to anybody. Run it locally. That's the beauty of Open Source. Another alternative to LM studio is Olama. Another fantastic product just a little bit more technical cuz it doesn't really come with a built-in interface. You have to install one yourself. 

So that's it. Those are the ways to run DeepSeek R1 right now. Definitely try it out. It is an insanely good model. I'm so excited for the future of Open Source.

Note, that there may be slight differences in text and video. Watch video for the original content.

DeepSeek R1 Video