DeepSeek-R1 is an open-source LLM (language model) optimized for reasoning and problem-solving. We’ll use the distilled 1.5B-parameter version (“DeepSeek-R1-Distill-Qwen-1.5B”), which is the smallest CPU-friendly model (~1.1 GB disk). This guide shows how to install and run DeepSeek on Ubuntu or RHEL/CentOS using the simplest methods (Docker or prebuilt binaries). You will learn how to download the model and run a prompt as well as verifying output.

Prerequisites

• Operating System: A distribution of Linux (Ubuntu/Debian or RHEL/CentOS). (Debian/Ubuntu users may find some steps easier.)
• Memory & Storage: At least 8–16 GB RAM is recommended (the 1.5B model uses ~7 GB RAM, but 16 GB ensures smooth operation). Ensure at least 10–20 GB free disk space for the model and dependencies.
• Software:
• For Docker method: Install Docker (or Podman). Ensure the Docker service can run containers.
• For Ollama method (no Docker): You need curl (for downloading the installer) and a Linux shell. Install curl (sudo apt install curl on Ubuntu); if not present.

Method 1: Using Docker

1. Install Docker Engine:

• Ubuntu/Debian: $ sudo apt update $ sudo apt install -y docker.io You should see output indicating Docker installed, for example: Setting up docker.io (20.x.x) ...
• RHEL/CentOS: $ sudo yum install -y docker After installation, verify Docker is available: $ docker --version Docker version 24.0.x, build abcdef (Sample output: Docker version and build number.)
• Start Docker: $ sudo systemctl start docker $ sudo systemctl enable docker $ sudo systemctl status docker Ensure the service is “active (running)”. (If not active, start it. Checking service status is similar to verifying any service.)

1. Run the Ollama container: Ollama provides a Docker image with the CLI pre-installed. Launch it in the background with port 11434 exposed:

   $ sudo docker run -d --name deepseek-ollama -p 11434:11434 ollama/ollama

This outputs a container ID, e.g. tak12290s.... You can check it is running via docker ps.

3. Download (pull) the DeepSeek model: Inside the running container, use Ollama to pull the 1.5B model.

   $ sudo docker exec -it deepseek-ollama ollama pull deepseek-r1:1.5b

Sample output (may take a few minutes):

   Downloading deepseek-r1:1.5b (1.1 GB)...
   [#####################################] 100% 1.1G/1.1G
   Download complete. 

This command saves the model files in the container (and to the mapped volume, if any).

4. Run DeepSeek: Still inside the container, start the model and enter an interactive prompt. For example:

   $ sudo docker exec -it deepseek-ollama ollama run deepseek-r1:1.5b

The CLI will load the model. You will see a prompt like:

   DeepSeek-R1 (1.5B) is ready. Type your question below.
   > What is 2+2?
   4
   >

(The model responds with 4, as expected.) You can type any question after the > prompt. Press Ctrl+C or type exit to stop the session.

5. Stop the container (optional): When done, exit the Ollama session and stop Docker:

   $ sudo docker stop deepseek-ollama

(This will shut down the container.)

Method 2: Using the Ollama CLI (no Docker)

1. Install Ollama: Run the below in your terminal to install Ollama:

   $ curl -fsSL https://ollama.com/install.sh | sh

This downloads and executes the Ollama installer. You might see output like:

   Installing Ollama v0.X.Y...
   Ollama installed successfully.

Then verify the installation:

   $ ollama --version
   ollama version 0.X.Y

(You should see the version number of Ollama.)

2. Download and run DeepSeek: Use the Ollama commands to pull and run the model:

   $ ollama pull deepseek-r1:1.5b

Sample output:

   Downloading deepseek-r1:1.5b (1.1 GB)...
   [#####################################] 100% 1.1G/1.1G
   Download complete.

Then start the model (this opens an interactive prompt):

   $ ollama run deepseek-r1:1.5b

You’ll see a prompt where you can type questions:

   DeepSeek-R1 (1.5B) is ready. Enter your query:
   > What is the capital city of Portugal?
   Lisbon.
   >

The model answers (in this case, “Lisbon.”). Press Ctrl+C when you’re done to exit the chat.

3. List and remove models (free space): You can see which models are installed with:

   $ ollama list

Sample output:

   NAME              SIZE
   deepseek-r1:1.5b  1.1GB

To delete the model and free up space:

   $ ollama rm deepseek-r1:1.5b

Sample output:

   Removed model: deepseek-r1:1.5b

Troubleshooting + Tips

• Command not found: Ensure to ran the installer or start the Docker service. You may need to re-open your terminal or add Ollama to your PATH.
• Service not running: Start the Docker or Ollama service: e.g. sudo systemctl start docker or (for Ollama service) sudo systemctl start ollama.service.
• Insufficient memory: The model uses several GB of RAM, make sure you have at least 8 to 16 GB RAM free or try swapping. (Smaller models may reqquire less memory.)
• Performance tuning: Ollama uses all CPU threads by default setting. This can be restricted by setting the OLLAMA_NUM_THREADS environment variable. e.g export OLLAMA_NUM_THREADS=4 will use 4 threads. In case the model is running too slow or your system gets overloaded, this will help.
• Container issues: If you are using Docker, check running containers (sudo docker ps). Make sure that the container name of (deepseek-ollama) matches what you used. If you change the docker run parameters, you may need to remove the old container first (sudo docker rm deepseek-ollama).

You should now have DeepSeek running locally. Simply rerun ollama run deepseek-r1:1.5b (or the Docker exec command) anytime you want to chat with the model. This setup works entirely offline and lets you verify each step with the sample outputs above.