Run DeepSeek R1 Locally for Free with Ollama and OpenWebUI

Awhile back I wrote about how you can run your own local ChatGPT experience for free using Ollama and OpenWebUI with support for LLMs like DeepSeek R1, Llama3, Microsoft Phi, Mistral and more! With the recent, open source release of DeepSeek R1, it’s also supported to run locally with Ollama too! This article will take you through the steps to do this. If you’re looking for an intro to getting started with Ollama on your local machine, I recommend you read my “Run Your Own Local, Private, ChatGPT-like AI Experience with Ollama and OpenWebUI” article first, then come back here.

Why Run DeepSeek R1 Locally?

DeepSeek R1 is a powerful and efficient open-source large language model (LLM) that offers state-of-the-art reasoning, problem-solving, and coding abilities. Running it locally provides several advantages:

• Privacy: No data is sent to external servers, ensuring complete control over your interactions.
• Performance: Get faster responses by leveraging your local hardware rather than relying on cloud-based APIs.
• Cost-Efficiency: Avoid ongoing API costs associated with cloud-based AI services.
• Customization: Fine-tune and integrate the model into your specific workflows without third-party limitations.

Prerequisites

Before installing DeepSeek R1, ensure you have the following:

• Windows, macOS, or Linux operating system
• At least 16GB RAM (for smaller models) and more for larger variants
• Ollama installed on your system

If you haven’t installed Ollama yet, you can download it from Ollama’s official website and follow their installation instructions.

Installing and Running DeepSeek R1 with Ollama

Step 1: Install Ollama on Windows and macOS

If you haven’t already installed Ollama, follow these steps:

1. Download and install Ollama from the official website.
2. Once installed, verify the installation with the command: ollama --version

Step 2: Download the DeepSeek R1 Model

To pull the DeepSeek R1 model using Ollama, run the following command in your terminal:

ollama pull deepseek-r1

This will automatically download the DeepSeek R1 model and default to the 7B parameter size to your local machine. The download is several GB and download time depends on your internet speed.

Which size model should you download?

DeepSeek R1 is available in multiple sizes, each defined by the number of parameters. The more parameters a model has, the more powerful it is—but it also requires more memory and computational power.

Parameters in a large language model (LLM) refer to the internal values that influence the model’s predictions and responses. More parameters generally result in better reasoning, problem-solving, and contextual understanding, but they also demand more RAM and processing power.

Choosing the Right Model for Your Hardware

Below is a table summarizing the different DeepSeek R1 models, their hardware requirements, and their ideal use cases.

Most likely the largest size of the DeepSeek R1 model that you’ll be able to run locally will be the 14B or 32B model depending on your hardware. However, most people will likely be able to run the 7B or 14B model. It’s worth testing a couple different sizes to find the largest model you can run that will return responses in a short enough time to be acceptable for use.

Here’s some tips for choosing the model that will run best on your hardware:

• If you have limited RAM (8GB–16GB) → Use DeepSeek R1-1.3B or 7B for basic tasks.
• If you have at least 24GB RAM → DeepSeek R1-14B offers a strong balance of performance and usability.
• If you want the best performance and have 64GB+ RAM (or a high-end GPU) → Choose DeepSeek R1-32B for top-tier AI capabilities.
• If you want to run DeepSeek R1-70B or 671B, then you will need some seriously large hardware, like that found in data centers and cloud providers like Microsoft Azure and AWS.

Once you have chosen the DeepSeek R1 model size to run, here are the Ollama commands to download the model sizes:

# DeepSeek R1 1.3B
ollama pull deepseek-r1:1.3b

# DeepSeek R1 7B
ollama pull deepseek-r1:7b

# DeepSeek R1 14B
ollama pull deepseek-r1:14b

# DeepSeek R1 32B
ollama pull deepseek-r1:32b

GPU Acceleration

If your system has a dedicated GPU / graphics card, you can significantly improve model inference speed by utilizing GPU acceleration with Ollama. You will want to make sure you have a supported GPU and the latest graphics drivers for your card installed.

Here are the Windows requirements you’ll need to meet for Ollama to utilize your GPU:

• NVidia GPU: NVidia driver 452.39 or newer
• AMD Radeon GPU: Latest AMD Radeon Driver

Step 3: Verify Installation

To ensure the model was downloaded successfully, run:

ollama list

If installed correctly, you should see deepseek-r1 in the list of available models.

Step 4: Running DeepSeek R1 Locally

Once downloaded, you can run the model locally with:

ollama run deepseek-r1

This starts the model, allowing you to interact with it in your terminal.

You also don’t need to run the ollama pull command first, if you just run ollama run it will download the model then run it immediately.

Here’s the ollama run commands for the different DeepSeek R1 model sizes:

# DeepSeek R1 1.3B
ollama run deepseek-r1:1.3b

# DeepSeek R1 7B
ollama run deepseek-r1:7b

# DeepSeek R1 14B
ollama run deepseek-r1:14b

# DeepSeek R1 32B
ollama run deepseek-r1:32b

Step 5: Ask DeepSeek R1 Questions

Now that you have the DeepSeek R1 model running on your local machine, you can start asking it questions!

🎉 Congratulations, you are now running your own local DeepSeek R1 LLM!

Optional: Using OpenWebUI for a GUI Experience

If you prefer a graphical interface instead of using the terminal, you can pair Ollama with OpenWebUI:

1. Install Docker if you haven’t already.
2. Run the OpenWebUI Docker container: docker run -d -p 3000:3000 --name openwebui openwebui/openwebui
3. Access OpenWebUI at http://localhost:3000 and configure it to use Ollama as the backend.

Note: Keep in mind this is a local instance of OpenWebUI. Although, since this uses Docker, it is possible to host OpenWebUI on a server in the cloud too, if you want to make it available from other machines.

Fine-Tuning and Customization

For advanced users, you may want to fine-tune DeepSeek R1 for specific tasks. Ollama allows you to create custom models based on DeepSeek R1 by modifying prompt templates and response behaviors.

ollama create deepseek-custom --base deepseek-r1 --modify-config

Follow the prompts to configure your custom AI assistant.

Conclusion

With Ollama, running DeepSeek R1 locally is simple and offers a powerful, private, and cost-effective AI experience. Whether you’re a developer, researcher, or enthusiast, having access to a cutting-edge model like DeepSeek R1 on your local machine opens up endless possibilities.

By running DeepSeek R1 locally, you not only enhance privacy and security but also gain full control over AI interactions without the requirement of cloud services. This setup is particularly beneficial for enterprises looking to integrate AI into their internal systems, researchers requiring offline capabilities, and developers interested in experimenting with AI models efficiently. Furthermore, the combination of DeepSeek R1 and Ollama allows users to create highly customized AI applications tailored to specific needs.

As AI continues to evolve, the ability to run sophisticated models locally will become an increasingly valuable asset. Whether you’re exploring AI for personal use, professional development, or business applications, DeepSeek R1 provides a robust and accessible solution. Try it today and take your AI experiments to the next level!

Original Article Source: Run DeepSeek R1 Locally for Free with Ollama and OpenWebUI written by Chris Pietschmann (If you're reading this somewhere other than Build5Nines.com, it was republished without permission.)