Guide to Installing Ollama and Running DeepSeek-R1 Model (600+B)

Introduction

Ollama is a lightweight, extensible framework designed to run large language models (LLMs) locally, offering a simple API to create, manage, and interact with models. It supports a variety of models, including DeepSeek-R1, a reasoning-focused model developed by DeepSeek, which is noted for performance comparable to advanced models like OpenAI’s o1. The user’s reference to “r1 model 600+B” likely indicates the DeepSeek-R1 model with 671 billion parameters, as it exceeds 600 billion and is the largest variant available. This guide provides detailed steps to install Ollama on Linux and run the DeepSeek-R1 model, addressing hardware requirements and alternatives for less powerful systems.

Step 1: Installing Ollama on Linux

Ollama provides multiple installation methods for Linux, with the automatic installation script being the most straightforward. Below are the steps and options:

Automatic Installation

Run the following command in your terminal to download and install Ollama:

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

This script handles the download and setup process, installing Ollama to /usr/bin/ollama. After installation, verify it by checking the version:

ollama -v

Manual Installation (Optional)

For more control, you can install Ollama manually:

1. Download the binary:

   curl -L https://ollama.com/download/ollama-linux-amd64.tgz -o ollama-linux-amd64.tgz

2. Extract and install:

   sudo tar -C /usr -xzf ollama-linux-amd64.tgz

3. Start the server:

   ollama serve

4. Verify installation:

   ollama -v

Setting Up as a System Service

To ensure Ollama runs automatically on boot, configure it as a systemd service:

1. Create a user for Ollama:

   sudo useradd -r -s /bin/false -U -m -d /usr/share/ollama ollama
   sudo usermod -a -G ollama $(whoami)

2. Create the service file /etc/systemd/system/ollama.service with:

   [Unit]
   Description=Ollama Service
   After=network-online.target
   
   [Service]
   ExecStart=/usr/bin/ollama serve
   User=ollama
   Group=ollama
   Restart=always
   RestartSec=3
   Environment="PATH=$PATH"
   
   [Install]
   WantedBy=multi-user.target

3. Enable and start the service:

   sudo systemctl daemon-reload
   sudo systemctl enable ollama
   sudo systemctl start ollama

4. Check the service status:

   sudo systemctl status ollama

Hardware-Specific Installations

• AMD GPU: Download the ROCm version:

  curl -L https://ollama.com/download/ollama-linux-amd64-rocm.tgz -o ollama-linux-amd64-rocm.tgz
  sudo tar -C /usr -xzf ollama-linux-amd64-rocm.tgz

  Ensure AMD ROCm drivers are installed from AMD Drivers.
• ARM64 Systems: Use the ARM64 binary:

  curl -L https://ollama.com/download/ollama-linux-arm64.tgz -o ollama-linux-arm64.tgz
  sudo tar -C /usr -xzf ollama-linux-arm64.tgz

Step 2: Starting the Ollama Server

To make models available for use, start the Ollama server:

ollama serve

This runs Ollama in the background, enabling model execution via terminal commands or the Ollama API.

Step 3: Running the DeepSeek-R1 Model

The DeepSeek-R1 model is available in multiple sizes, with the 671B parameter version being the largest and likely the “600+B” model referenced. Below are the details for running it and alternatives for different hardware capabilities.

DeepSeek-R1:671B (Full Model)

To run the full DeepSeek-R1 model with 671 billion parameters, use:

ollama run deepseek-r1:671b

Hardware Requirements

• Size: Approximately 404 GB (4-bit quantized).
• VRAM: Over 400 GB, typically requiring at least 6 NVIDIA H100 GPUs or equivalent.
• RAM: Likely hundreds of GB, far exceeding typical consumer hardware.
• Storage: Ensure sufficient disk space (at least 404 GB free).

Challenges

Running the 671B model is resource-intensive and may not be feasible for most users without access to high-end server hardware, such as an EPYC server or a multi-GPU cloud setup. For example, a Digital Spaceport guide describes running it on a $2000 EPYC server, indicating the scale of hardware needed.

Alternative: Distilled DeepSeek-R1 Models

Ollama offers smaller, distilled versions of DeepSeek-R1, fine-tuned from the 671B model using Llama and Qwen architectures. These are more practical for standard hardware:

Model Command	Parameters	Size	Minimum RAM
ollama run deepseek-r1:1.5b	1.5B	1.0 GB	4 GB
ollama run deepseek-r1:7b	7B	4.7 GB	8 GB
ollama run deepseek-r1:14b	14B	9.1 GB	16 GB
ollama run deepseek-r1:32b	32B	20 GB	32 GB
ollama run deepseek-r1:70b	70B	43 GB	64 GB

These models maintain strong reasoning capabilities and are suitable for users with consumer-grade hardware. For example, the 70B model is noted to outperform some closed-source models like OpenAI’s o1-mini on benchmarks (Ori Cloud Guide).

Quantized Versions

Quantized versions of the 671B model reduce memory requirements, making them accessible on less powerful hardware:

• 1.58-bit Quantized Model:

  ollama run SIGJNF/deepseek-r1-671b-1.58bit

  • Can run on a single NVIDIA 4090 GPU (Ollama Quantized Model).
  • Maintains much of the 671B model’s performance with a smaller footprint.

• 1.73-bit Quantized Model:

  ollama run Huzderu/deepseek-r1-671b-1.73bit

  • Similarly optimized for single-GPU setups (Ollama Quantized Model).

These quantized models are ideal for users who want to experiment with the 671B model’s capabilities without requiring extensive hardware.

Step 4: Interacting with the Model

Once a model is running, you can interact with it in two ways:

Terminal Interaction

After running ollama run deepseek-r1:671b (or another variant), you can type prompts directly in the terminal to interact with the model.

API Interaction

To integrate the model into applications, use the Ollama API. For example, send a prompt using curl:

curl http://localhost:11434/api/chat -d '{
  "model": "deepseek-r1:671b",
  "messages": [{"role": "user", "content": "Solve: 25 * 25"}],
  "stream": false
}'

This allows programmatic access to the model’s capabilities (DataCamp Tutorial).

Additional Considerations

• Model Licensing: DeepSeek-R1 models are licensed under the MIT License, supporting commercial use and modifications. Distilled models may have additional licensing considerations (e.g., Qwen models under Apache 2.0, Llama under its own license).
• Updates: To update Ollama, re-run the installation script or download the latest binary.
• Logs: View service logs with:

  journalctl -e -u ollama

• Uninstallation: To remove Ollama:

  sudo systemctl stop ollama
  sudo systemctl disable ollama
  sudo rm /etc/systemd/system/ollama.service
  sudo rm $(which ollama)
  sudo rm -r /usr/share/ollama
  sudo userdel ollama
  sudo groupdel ollama
  sudo rm -rf /usr/local/lib/ollama

Conclusion

Installing Ollama on Linux is straightforward using the provided script, enabling you to run powerful models like DeepSeek-R1. The 671B parameter model, likely the “600+B” model referenced, offers exceptional reasoning capabilities but requires significant hardware, such as multiple high-end GPUs. For most users, distilled models (e.g., 70B or 7B) or quantized versions (e.g., 1.58-bit on a single 4090) provide a practical alternative. By following the steps above, you can set up Ollama and choose the DeepSeek-R1 variant that best fits your hardware and needs.

For further details, explore the Ollama Library or refer to community guides like the DataCamp Tutorial.