With AI becoming mainstream among most websites, AI programs have become a necessity for almost everyone. Now that the industry-disrupting new sensation from China’s DeepSeek is here, what better way to reap the benefits than self-hosting it?
Now, while it’s an open-source AI model, you can self-host DeepSeek locally on your own device, you will need a PC with a high-specs build to run it properly. In such cases, you can host it on a cloud server to ensure maximum benefits.
Why Self Host DeepSeek?
Firstly, you may be wondering why you may want to self-hosting DeepSeek AI in the first place. Here are some reasons worth considering.
Privacy: DeepSeek AI is hosted on Chinese servers, which will store your information there to train their models. By hosting it in your own infrastructure, you will not have to worry about your data being utilized by a foreign company.
Efficiency: With lower latencies granted for self-hosting, it will improve your usage efficiency, especially with smaller parameters.
Scalability: Depending on what type of model you will be using, you can scale up or down based on your preferences.
Control: Since you will be hosting on your own infrastructure, you will be less reliant on 3rd parties, giving you full control over your AI model.
Exploring DeepSeek
You may be curious as to why DeepSeek is making waves in the industry. Sure, it is much more efficient to build compared to other generative AI available, but how does it affect the user? The answer lies in their new features that drastically change how gen AI works. Learn more about them in the following slides.
The beauty of DeepSeek AI and it being open source is that due to the availability of various models, you can scale up or down your parameters based on your needs and resources. Not to mention, you are going to get very comparable results to something like ChatGPT even on lower parameters.
Step-by-step guide on how to host DeepSeek AI on the cloud
For this tutorial, we will be hosting our AI on a Ubuntu Cloud server on a Linux OS. Don’t fret if you are using a different cloud server,as the steps will be pretty much the same. We will be utilizing Ollama, a command line tool that can help you host DeepSeek in a very easy way.
Step 1 – Login to your Cloud Server
For the first step, log in to your cloud server. As mentioned before, we used an Ubuntu Cloud Server to self-host DeepSeek.
Step 2 – Download Ollama
Go to Ollama.com and download it from the site. It is available for macOS, Linux, and Windows. Make sure you download the version based on the OS you are using with your cloud server.
For this demonstration, we are using Ubuntu running on Linux OS. Once you click on the Linux button, you will be presented with a link. Copy the link and paste it on your server.
Now wait till Ollama is installed onto your server.
Step 3 – Pick your model
Once installed, you should select the DeepSeek model. Go back to the Ollama site and click on “Models.”
You will see a list of various models available to use. Select deepSeek-r1.
Now it is time to pick the model with the most suitable parameters. Due to DeepSeek’s knowledge distribution technique, smaller models can imitate larger models to provide very comparable results. This makes it so that even servers or hardware of lower specs can utilize and host AI models. For the purpose of this guide, we will demonstrate the process with a model with parameters of 1.5B.
Once you choose deepseek r-1, you will be able to pick your parameters. Click on the parameter options list as shown and click on 1.5B.
Step 4 – Download the selected model onto your Cloud server
All you have left to do is download the model onto your server. First, copy the link of the selected model as shown.
The model will start to download once you paste it onto the server terminal.
It will take a moment for the program to be downloaded.
Step 5 – Enjoy!
Now, you can fully utilize DeepSeek generative AI on your server. Just add some prompts to see the results.
We hope this guide helps you self-host your own AI model so that you do not have to worry about privacy, and can operate on your cloud infrastructure.
Hardware requirements
Before we get on to the steps required to host DeepSeek R1, let us have a look at what kind of hardware you need to be able to run it. Here is a formula that can help you determine that based on your LLM.
M = ((P x 4B)/(32/Q)) x 1.2
Here,
M = Required VRAM of GPU
P = Number of parametre in model
4B = 4 bytes per parametre (Deepseek uses FP8 precision)
Q = Number of bits per parametre when loading
1.2= A 20% overhead factor
For our purposes, we will be showing how to use a 1.5B model. Mind you DeepSeek uses FP8 precision. So, our required specifications to run it properly can be calculated using the given formula.
Here,
B= 1,500,000,000
Q= 8
M = (1,500,000,000 x 4)/(32/8)) x 1.2
= 1.8 GB
So, to run a 1.5B DeepSeek r1 model, you will need about 1.8 GB of GPU memory. You will either need hardware with such specifications. In our case, make sure your cloud server can handle that kind of requirement.
Challenges you may face
Of course, with self-hosting any programs, you will face some difficulties along the way. These are some of the things you need to be aware of before you dive in to this project.
Biasness and censorship: Most generative AI models have restrictions and censorship on controversial topics. While it will useful for general harmless tasks, be sure to avoid such topics. It should be noted that there are uncensored models available for you to utilize.
Cost and maintenance: Self-hosting means you will need to incur a high initial investment to get things started. You will also need to conduct maintenance regularly, which will add up to a lot of ongoing expenses.
Scalability: Scaling up your model will require resources on your end, adding up your maintenance costs.
With all that out of the way, let’s look at how to self-host DeepSeek AI on cloud platforms.
More to Explore
The Quest for Knowledge Continues. Fuel Your Curiosity.
