The future is here! A detailed guide to local deployment of domestic AI - DeepSeek!

Preface

Hello everyone, I am Andy

I have been deeply involved in the cross-border field for 6 years and am good at e-commerce + technology application to explore the blue ocean market. I will share a lot of cross-border overseas and technology-related analysis. If you are interested in this direction, you can follow me. For some content that is not convenient to shoot into a video, I will write it into a graphic SOP to make it more convenient for everyone to operate.

The hottest technology news during the Spring Festival should be Deepseek. Feng Ji, CEO of Black Myth Wukong Game Science, even directly called it

A national-level technological breakthrough!

I didn’t pay much attention to it at the beginning, but I play Heihou a lot, and I believe that Mr. Feng is also a serious person. It must be worth paying attention to to get his trust endorsement, so I started to learn about the relevant information: digesting the core content of Douyin, Zhihu, and Bilibili.

Later I found out that Liang Wenfeng, the founder of Huanfang, is also from Wuchuan. As a fellow villager, I feel proud to meet him. I dug deeper into the founder's growth experience, past sharing, and values. I couldn't help but sigh: small town test-takers are truly China's unique wealth. At the same time, I really have to believe in the metaphysics of "national destiny."

The direction that OpenAI (CloseAI) is leading now is to use computing power equipment to achieve miracles, while at the same time imposing export controls on high-end graphics cards in order to restrict the development of domestic AI technology. However, there are always idealistic geniuses who can find alternative ways to break through and optimize training costs through algorithms. From training costs to API prices, we can see extreme cost control, and see that AI applications can really become applications of a universal nature like water and electricity, and technology can change lives.

I have shared the content of ChatGPT on the forum quite early, but I have to say that there is still a huge threshold. I am willing to delve into it because of my interest + industry, but for ordinary people, it takes network configuration (scientific) + account registration + SMS code reception + anti-feng number processing to basically discourage more than 90% of the people.

Now the threshold has been lowered. With just an 86 mobile phone number, you can get this kind of magic weapon. I think this kind of universal AGI is just like the omniscient and omnipotent "sage" that people in ancient times believed in. And a sage who can respond to any request. I recommend that you also think about how to combine the industry for implementation and increase the frequency as much as possible (there is also a kind of anxiety: non-core jobs are too replaceable, especially in the e-commerce field, so you must be sensitive enough and actively embrace changes)

Because Deepseek is so popular now, its server crashed and it also received a lot of malicious attacks. The API application entry was also closed . However, we can see the genes of a great company in DeepSeek: it has this core technology (which can bring huge economic returns), but is willing to open source it directly.

It has to be said that Huanfang is still idealistic. The company is a top domestic quantitative private equity institution and must have evaluated the commercial economic returns of the DS project. It has given up huge profits, but it has a huge positive boost to the development of the global AI ecosystem. It used to require huge financial resources to be invested, but after open source, everyone has the opportunity to sit at the table.

I have shared some basic information above. Every time I talk about the core content, I will talk about the background (doge)


What I will share today is: DeepSeek local deployment (technical) local deployment solution

I will have the opportunity to share with you relevant information at the application level later:
1. AI applications operated by Amazon (basic applications)
2. Establishing a knowledge base at the company level (macro application)

Open the Ollama official website and click download according to your system
https://ollama.com/


After downloading, just follow the default settings to complete the installation, then return to the Ollama homepage and click the model subpage
Ollama has aggregated some popular open source models. We directly find the DeepSeek-r1 model.

You can find that there are many different versions here, choose the one that matches your computer configuration (I bought 4090 immediately after the New Year)

For specific configuration, please refer to this table

Select the corresponding model and copy this code

Mac users open [Terminal] locally on their computer
Windows users open the [command line cmd] locally on the computer
Then Mac users open [Terminal] locally on their computer
Windows users open [cmd] locally on their computer
Paste the code you just copied, press Enter, and start downloading the model
After the download progress bar is finished, it will be installed automatically. After everything is completed, we can start using it directly in the terminal.

PS. Be sure to choose a model that matches your computer configuration. My Mac is equipped with a 7B, and the back panel temperature rises sharply.

After the installation is complete, we can use it directly in the terminal box, but it is definitely not very convenient in actual use. So we introduce two methods below to add an interactive interface to the local deployment, which is more in line with our usage habits

Open the chatbox official website and start the web version. Chatbox also provides clients to suit your usage habits.
https://chatboxai.app/zh

Then click [Settings] in the lower left corner

Model provider: Ollama

Then follow this tutorial to set up
https://chatboxai.app/zh/help-center/connect-chatbox-remote-ollama-service-guide

You can see the tutorial for the corresponding system.

Then go back to the official website of Chatbox, click [Settings], select the model we downloaded, click Save, and you can start using it. PS: If the model is not displayed here, it is probably because you forgot to reopen Ollama.

Chatbox also provides the option to create an agent that can help us handle specific tasks: such as listing optimization

I will update the Docker solution when I have time later. Docker has more ways to play and can create voice intelligent agents, but it is relatively more complicated. The Chatbox above can already meet the usage needs of most people. I will decide whether to update this part based on everyone's feedback.