How to Run DeepSeek R1 Locally (4 Simple Steps)

January 2025, Deepseek took the world by storm. Aside from being a lot lighter and a LLM that can be used locally, it supposedly only cost 6 million to develop!

Why is that noteworthy? It’s mindblowing since ChatGPT is estimated to have cost around 1 billion USD to develop (which we’ll dive into deeper later).

But if you’re looking to run an LLM locally, and are keen to know more about Deepseek - we’ll dive into that more below.

What is DeepSeek?

Started by Liang Wenfeng and owned by the Chinese hedge fund High-Flyer, DeepSeek is an advanced language model designed to handle a wide range of natural language processing tasks.

Contrary to popular belief, SemiAnalyisis, an independent research firm debunked the cost of DeepSeek, pointing out that the 6 million is likely only the cost of the GPU pre-training and does not consider other capital expenditures!

This AI model excels in generating human-like responses, understanding context, and adapting to various use cases. Built with a focus on efficiency, DeepSeek aims to support users in research, development, and automation

What is DeepSeek R1?

DeepSeek R1 is a specific version of the DeepSeek language model optimized for local deployment. It is designed to run efficiently on personal systems while maintaining high-quality performance

This is one of the latest models released in January 2025 and can be used for text generation, summaries to understand things, and to develop code.

How to Run DeepSeek R1 Locally? 

Step 1: Download and Install Ollama

Once installed, verify the setup by running a simple command to check if Ollama is functioning correctly.

Step 2: Add Deepseek to Your System Using Ollama

After installing Ollama, the next step is to integrate DeepSeek into your local environment. This involves fetching the DeepSeek model and configuring it within Ollama through the command:

ollama pull deepseek-r1

Step 3: Use Ollama Serve

Ollama Serve allows users to interact with DeepSeek in real-time. By running the serve command, you initiate the model and make it accessible for generating responses by typing:

ollama serve

Step 4: DeepSeek is Ready to Use

When installed go ahead and type the command:

ollama run deepseek-r1

How is Deepseek Different from ChatGPT?

While ChatGPT is known for conversational depth, DeepSeek is designed for specific tasks that demand precision and adaptability. The choice between the two depends on the user’s goals and workload requirements.

Deepseek Vs. ChatGPT: Comparison

Criteria

ChatGPT

DeepSeek

Response Accuracy and Relevance

Generally strong; varies by task

Rapidly improving; focus on code

APIs and plugins

Extensive API access; growing plugins

Not as many

Customization and Fine-Tuning 

Limited fine-tuning options

Open models enable customization

Cost and Licensing Model 

Usage-based pricing; commercial licenses

Open source; potentially lower cost

Model Architecture

Proprietary; transformer-based

Mixture of experts; transformer

Deepseek Large Language Models (LLMs)

We’ve gone ahead and listed the different DeepSeek LLMs in chronological order.

1. DeepSeek Coder (November 2023): This marked DeepSeek's initial foray into open-source coding-focused models.  It was designed to assist with various coding tasks, demonstrating their commitment to the developer community. This release was significant as it allowed developers to experiment with and build upon DeepSeek's code generation capabilities.
2. DeepSeek LLM (December 2023): This was the first iteration of DeepSeek's general-purpose language model. It laid the foundation for their subsequent LLM development. This model was a starting point, showcasing their ability to train a foundational language model.
   
3. DeepSeek-V2 (May 2024): This version of their LLM emphasized improved performance and reduced training costs compared to the initial LLM.  This highlights DeepSeek's focus on efficiency and making large language models more accessible.  It signifies a step forward in model optimization.
   
4. DeepSeek-Coder-V2 (July 2024):  This model, with its 236 billion parameters and extended 128,000 token context window, was designed to tackle more complex coding challenges.  The larger context window is particularly noteworthy, enabling the model to handle more extensive codebases.  This demonstrates a focus on practical, real-world coding applications.
   
5. DeepSeek-V3 (December 2024): This model introduced a mixture-of-experts (MoE) architecture and boasted 671 billion parameters with a 128,000 token context length. The MoE architecture is significant for its potential to improve model capacity and efficiency.  This signifies a move towards more advanced and powerful model architectures.
   
6. DeepSeek-R1 (January 2025): Building upon DeepSeek-V3, this model focused on advanced reasoning capabilities.  DeepSeek aimed for competitive performance with leading models like OpenAI's GPT-4 (referred to as "o1" in your prompt, likely a typo) while maintaining lower costs.  This emphasizes DeepSeek's goal of achieving high performance with greater efficiency.
   
7. Janus-Pro-7B (January 2025): This vision model demonstrates DeepSeek's expansion into multimodal AI. Janus-Pro-7B can understand and generate images, opening up new possibilities for applications.  This signifies DeepSeek's research and development in the exciting field of vision-language models.

What is Deepseek Used For?

DeepSeek is used across the board including in research, content generation, and automation. For developers, using DeepSeek in applications requiring natural language understanding can be one way to go. However, for businesses, customer support and data analysis are more likely uses.

FAQs About DeepSeek

What are the system requirements for running DeepSeek locally?

DeepSeek requires a system with sufficient processing power, memory, and storage. While exact specifications depend on the model version, a modern CPU and GPU help achieve optimal performance.

Can DeepSeek be used offline?

Yes, once installed, DeepSeek can function without an internet connection. However, periodic updates may require online access.

How does DeepSeek compare to other AI models?

A: DeepSeek focuses on efficiency and precision, offering a streamlined approach to text generation. While some models prioritize conversational engagement, DeepSeek is designed for tasks requiring clarity and accuracy.

Is DeepSeek free to use?

Availability and pricing depend on the version and intended use. Some models may be accessible for free, while advanced features could require a subscription or licensing.