{"id":27856,"date":"2025-01-26T10:03:58","date_gmt":"2025-01-26T02:03:58","guid":{"rendered":"https:\/\/www.1ai.net\/?p=27856"},"modified":"2025-02-06T11:51:36","modified_gmt":"2025-02-06T03:51:36","slug":"%e5%a6%82%e4%bd%95%e5%9c%a8%e4%bd%a0%e8%87%aa%e5%b7%b1%e7%94%b5%e8%84%91%e4%b8%8a%e5%ae%89%e8%a3%85%e5%92%8c%e4%bd%bf%e7%94%a8deepseek-r1-%e5%a4%a7%e6%a8%a1%e5%9e%8b%ef%bc%9fdeepseek-r1%e6%9c%ac","status":"publish","type":"post","link":"https:\/\/www.1ai.net\/en\/27856.html","title":{"rendered":"How to install and use DeepSeek R1 Big Model on your own computer?DeepSeek R1 Local Deployment Guide"},"content":{"rendered":"

Everyone seems to be talking aboutDeepSeek<\/a> R1, a brand-new AI company developed by the domestic AI company DeepSeekOpen Source AI Language Model<\/a>.<\/p>\n

Some users claim that it is on par, if not better, than OpenAI's o1 in terms of reasoning power.<\/p>\n

\"b72a7bf5j00sqoa6u0026d000u000hwp\"<\/p>\n

Currently, DeepSeek is free, which is good news for users, but it also raises some questions.<\/p>\n

How are they managing server costs as their user base surges?<\/p>\n

Hardware shouldn't be cheap to run, should it?<\/p>\n

A reasonable presumption is - data.<\/p>\n

Data is the lifeblood of AI models.<\/p>\n

They may collect user data in some way that may be beneficial to their quantitative trading models or used for other forms of realizations.<\/p>\n

So if you're concerned about data privacy but still want to use R1 without sharing the data, the best way to do that is to run the model locally.<\/p>\n

What is DeepSeek R-1?<\/strong><\/p>\n

A few days ago, DeepSeek R-1 was unveiled as a fully open-source model, meaning that anyone can access its underlying code, make modifications, and even fine-tune it to suit their needs.<\/p>\n

From a technical point of view, DeepSeek R1 (often abbreviated as R1) derives from a large base model called DeepSeek-V3.<\/p>\n

The research team optimized this model by combining supervised fine-tuning (SFT) and reinforcement learning (RL) with high-quality manually labeled data.<\/p>\n

The result is a chatbot that can handle complex prompts, reveal the reasoning steps behind complex problems (sometimes more transparently than other models), and even present code in the chat interface for quick testing.<\/p>\n

Honestly, really very impressive, especially for an open source model.<\/p>\n

How to run locally<\/strong><\/p>\n

To run DeepSeek R1 locally, we will use a tool called Ollama.<\/p>\n

Ollama is a free open source tool that allows users to run Large Language Models (LLMs) locally on their computers. It is available for macOS, Linux and Windows.<\/p>\n

Visit the official website of Ollama, click on the \"Download\" button and install it on your system.<\/p>\n

\"05388ab2j00sqoa7t000zd000tq00hkp\"<\/p>\n

To confirm that the installation was successful, open a terminal and run the following command:<\/p>\n

ollama-v<\/p>\n

\"ad3d9722j00sqoa9u0010d000ti00dwp\"<\/p>\n

You should see the version number of Ollama, not an error message.<\/p>\n

Under the \"Models\" tab, search for the keyword \"deepseek\" and you should see \"deepseek-r1\" in the first position of the search list.<\/p>\n

\"37d965adj00sqoaar001rd000m600lsp\"<\/p>\n

After clicking on it, in the \"Models\" section, you'll notice that there are multiple model sizes with parameter counts ranging from 5 billion to 671 billion. As a rule of thumb, larger models require more powerful GPUs to run.<\/p>\n

\"a7976b7cj00sqoacs001id000nv00hep\"<\/p>\n

Smaller models like the 800 million parameter version can run on GPUs with 8GB of video memory. Larger models, on the other hand, require significantly more resources (see the Graphics Memory and GPU Requirements section below).<\/p>\n

To download and run the model with 800 million parameters, use the following command:<\/p>\n

ollama run deepseek-r1:8b<\/p>\n

The model will begin to download (approximately 4.9GB). Make sure you have enough disk space before proceeding.<\/p>\n

\"e1d36abdj00sqoac60019d000to00dup\"<\/p>\n

Once the download is complete, the model will run on your local computer and you can talk to it immediately.<\/p>\n

\"5e12f8f2j00sqoadc001jd000tq00dup\"<\/p>\n

Let's test it with this sample prompt:<\/p>\n

\"fb2f4602j00sqoadt001vd000tr00eop\"<\/p>\n

\n
    \n
  • Tip: What is DeepSeek R-1?<\/li>\n
  • RESPONSE: DeepSeek-R-1 is an AI assistant developed by Chinese company DeepSeek. It aims to provide answers and help on a variety of topics, including but not limited to math, programming, natural language processing, and more. If you have any questions or need help, feel free to ask me!<\/li>\n<\/ul>\n<\/blockquote>\n

    That's great.<\/p>\n

    It runs fast and works even when I disconnect my laptop's Wi-Fi connection. Note that it can't access the web even if you are connected to the internet.<\/p>\n

    \"94893904j00sqoae5001sd000tr00eop\"<\/p>\n

    \n
      \n
    • FYI: What is AAPL's latest stock price?<\/li>\n
    • RESPONSE: As an AI assistant, I do not have access to real-time data and therefore cannot provide an up-to-date stock price for Apple Inc (AAPL). For the most accurate and up-to-date information, I suggest you check financial news platforms or your brokerage service.<\/li>\n<\/ul>\n<\/blockquote>\n

      Something else Ollama can do:<\/p>\n

        \n
      • Locally running LLMs, including LLaMA2, Phi 4, Mistral and Gemma 2<\/li>\n
      • Allow users to create and share their own LLMs<\/li>\n
      • Packaging model weights, configurations and data into a single package<\/li>\n
      • Optimized settings and configuration details, including GPU usage<\/li>\n<\/ul>\n

        GPU and Memory Requirements<\/p>\n

        The graphics memory requirements for DeepSeek-R1 depend on factors such as the size of the model, the number of parameters, and the quantization technique. Below is a detailed overview of the graphics memory requirements for DeepSeek-R1 and its refined model, as well as the recommended GPUs:<\/p>\n

        \"f6114070j00sqoawo008ad000v900qzp\"<\/p>\n

        Key notes on video memory usage:<\/p>\n

          \n
        • Distributed GPU setup for large models<\/strong>: Running DeepSeek-R1-Zero and DeepSeek-R1 requires a large amount of video memory, so distributed GPU configurations (e.g., NVIDIA A100 or H100 in a multi-GPU setup) are required for optimal performance.<\/li>\n
        • Single GPU compatibility for streamlined models<\/strong>: The lite model has been optimized to run on a single GPU with low video memory requirements, with a minimum requirement of just 0.7 GB.<\/li>\n
        • Additional memory usage<\/strong>: Activation, buffer and batch tasks may consume additional memory.<\/li>\n<\/ul>\n

          Why run locally?<\/strong><\/p>\n

          Of course, DeepSeek's web chatbot and mobile app are free and easy to use.<\/p>\n

          You don't have to do any setup, and features like DeepThink and web search are built in. But running it locally might be a better option for the following reasons:<\/p>\n

          private business<\/strong><\/p>\n

          When you use the web or app version, your query and any attached files are sent to DeepSeek's servers for processing.<\/p>\n

          What happens to this data?<\/p>\n

          We don't. Running the model locally ensures that your data stays on your computer, giving you complete control over your privacy.<\/p>\n

          offline access<\/strong><\/p>\n

          Running the model locally means you don't need an Internet connection.<\/p>\n

          If you're traveling, experiencing spotty Wi-Fi, or just prefer to work offline, the local setting lets you use DeepSeek anytime, anywhere.<\/p>\n

          future-proofing<\/strong><\/p>\n

          DeepSeek's service is currently free, but that's unlikely to last forever.<\/p>\n

          At some point, they may need to be monetized and usage restrictions or subscription fees may come into play. By running the model locally, you can avoid these limitations altogether.<\/p>\n

          flexibility<\/strong><\/p>\n

          When using the local version, you are not limited by the default settings.<\/p>\n

          Want to fine-tune your model? Integrate it with other tools? Build a customized interface? The open source nature of DeepSeek R1 opens up endless possibilities.<\/p>\n

          Summarize<\/strong><\/p>\n

          It is still unclear how DeepSeek handles user data. If you're not too worried about data privacy, using a web or mobile app may be a better option as they are easier to use and offer features like DeepThink and web search.<\/p>\n

          But if you care about where the data goes, running the model locally is a good alternative to consider.<\/p>\n

          DeepSeek models are designed to work well even when the hardware is not particularly powerful.<\/p>\n

          While large models like DeepSeek-R1-Zero require a distributed GPU setup, the lite version makes it possible to run smoothly on a single GPU with lower video memory requirements.<\/p>","protected":false},"excerpt":{"rendered":"

          Everyone seems to be talking about DeepSeek R1, the new open source AI language model developed by domestic AI company DeepSeek. Some users claim that it is on par or even better than OpenAI's o1 in terms of reasoning ability. Currently, DeepSeek is free, which is good news for users, but it also raises some questions. With the surge in users, how are they managing server costs? Hardware running costs shouldn't be cheap, right? A reasonable speculation is - data. Data is the lifeblood of AI models. They probably collect user data in some way that could be beneficial to their quantitative trading models or used for other forms of realizations. So, if you're concerned about data privacy, but still want to make<\/p>","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[144],"tags":[3606,5646,5645],"collection":[5669],"class_list":["post-27856","post","type-post","status-publish","format-standard","hentry","category-baike","tag-deepseek","tag-ai","tag-5645","collection-deepseek"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/27856","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/comments?post=27856"}],"version-history":[{"count":0,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/27856\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/media?parent=27856"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/categories?post=27856"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/tags?post=27856"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/collection?post=27856"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}