Speaker 1: Do you want the power of OpenAI's O1 reasoning model without the insane $60 per million token price tag? DeepSeek just dropped their R1 model which is an open source alternative to OpenAI's O1 reasoning model that you can run completely free on your own machine. I should mention you can also test it out through their website and their API actually costs a fraction of the OpenAI O1 model. In this video we'll focus on running the model locally on our own machines for free using OLAMA. So how are these reasoning models different to traditional large language models? Well most large language models are like quick thinking students who rush to give you the first answer that pops up in their head. But reasoning models are different. It's like having a thoughtful expert who shows their work and explains the reasoning step by step. This chain of thought process also means that the model can indeed correct itself and improve the final result. Let me give you a quick example of this. Let's ask chatgpt to plan a complex event. Like let's plan a wedding destination in Bali with a specific budget. Then also think through this step by step and consider things like what the essential costs are, what the potential problems are and how we can plan for them, how we can creatively maximize the budget and what's the minimum viable budget we'd need to work with or should we consider alternative destination. So I'm going to run this and do take note this is not using O1. This is just a standard gpt 4.0 model and we can see the response is streaming through very quickly which also means there's not a lot of thought and reasoning going into this process. It's simply spitting out the first thing it thought of. On the other hand let's have a look at what this looks like when we run this model in OLAMA and this is running locally on my own machine. It's saying the same prompt. First you'll see that the model is thinking through the process so it's kind of reasoning through the step by step which takes a bit longer to execute but what's fascinating is it's sort of correcting itself by saying wait let me think and it's kind of reasoning through the process as we go along. Now it's thinking about the food, it's thinking about decorations, it's thinking about transportation, emergencies and we can see it's sort of working out budgets and it's adding up all these different values and scrolling down even more. We can see that it's calculated the total minimum amount and finally it's completed its thought process and now it's providing us with that final answer. So yes it's a bit slower than using regular large language models but because it's reasoning through the problem the final result is just so much better making this perfect for working with code, math, complex puzzles and complex instructions like this that needs a bit of planning and thought put into it. Now let's have a look at running this model locally on our own machine. Head over to olama.com. OLAMA is a fantastic tool that makes it easy to run large language models on your own machine. It's completely free to use and super simple to set up. All you have to do is click on download then select your operating system then download the installer then simply run the installer and go through the setup process. Once OLAMA has installed you can see if it's up and running by opening your terminal or command prompt and typing OLAMA and if everything was set up correctly you should see this list of available commands. Now all we have to do is download the DeepSeq model. Back on the OLAMA website go to models and at the time of recording the DeepSeq model is at the top of the results but if you don't see it simply search for DeepSeq R1 then click on this result and from this page we can see that there are different model sizes starting from a 1.5 billion parameter model and all the way to a 671 billion parameter model. For commercial hardware and for the majority of you the 1.5 billion parameter model or the 7 billion parameter model will work perfectly fine. If you have more powerful hardware you can definitely try the 14 billion parameter model or the 32 billion parameter model or if you have a potato you can simply run the 1.5 billion parameter model. For this video let's use the 70 billion parameter model so from this drop down let's select it then on the right hand side simply copy this command back in your terminal simply run that command. This will download the DeepSeq R1 model and afterwards you'll be able to send it a message. Let's just say hello and we can see the thought process and because this was such a simple prompt it didn't have to think too hard on solving this problem and now we can give it a complex puzzle or math problem or event to plan and we can see its thought process and if we scroll down we can see the final result from this reasoning model and now that we have our model up and running we can integrate it with AI builders like N8n and Flowwise AI but there's a lot more we can do with Olama. We can provide system prompts to this model, we can create custom models from it and we can interact with this model from the Olama API. So to learn all the ins and outs using Olama check out this other video over here. I'll see you in the next one. Bye-bye.