Preface
With the rapid development of artificial intelligence technology, the demand for local deployment of large models is also increasing. As an open source and powerful large language model, DeepSeek provides flexible local deployment solutions, allowing users to run models efficiently in local environments while protecting data privacy. The following is a detailed DeepSeek local deployment process.
1. Environmental preparation
(I) Hardware requirements
- Minimum configuration: CPU (supports AVX2 instruction set) + 16GB memory + 30GB storage.
- Recommended configuration: NVIDIA GPU (RTX 3090 or higher) + 32GB memory + 50GB storage.
(II) Software dependency
- operating system: Windows, macOS or Linux.
- Docker: If you use the Open Web UI, you need to install Docker.
2. Install Ollama
Ollama is an open source tool for easy running and deploying large language models locally. Here are the steps to install Ollama:
- Visit Ollama official website: Go to Ollama's official website and click the "Download" button.
- Download the installation package: Select the corresponding installation package according to your operating system. After the download is completed, double-click the installation file and follow the prompts to complete the installation.
-
Verify installation: After the installation is completed, enter the following command in the terminal to check the Ollama version:If the version number is output (e.g.
ollama --version
ollama version is 0.5.6), it means the installation is successful.
3. Download and deploy the DeepSeek model
Ollama supports multiple DeepSeek model versions, and users can choose the appropriate model according to the hardware configuration. Here are the deployment steps:
Select a model version:
- Entry level: Version 1.5B, suitable for preliminary testing.
- Mid-range: 7B or 8B version, suitable for most consumer-grade GPUs.
- high performance: 14B, 32B or 70B version, suitable for high-end GPUs.
Download the model:
Open the terminal, enter the following command to download and run the DeepSeek model. For example, the command to download version 7B is:
ollama run deepseek-r1:7b
If you need to download another version, you can refer to the following command:
ollama run deepseek-r1:8b #8B versionollama run deepseek-r1:14b #14B versionollama run deepseek-r1:32b #32B version
Start the Ollama service:
Run the following command in the terminal to start the Ollama service:
ollama serve
After the service is started, you can interact with the model by visiting http://localhost:11434.
4. Use Open Web UI (optional)
To interact with the DeepSeek model more intuitively, the Open Web UI can be used. Here are the steps to install and use:
- Install Docker: Make sure Docker is installed on your machine.
-
Run Open Web UI：
Run the following command in the terminal to install and start the Open Web UI:
docker run -d -p 3000:8080 \ --add-host=:host-gateway \ -v open-webui:/app/backend/data \ --name open-webui \ --restart always \ /open-webui/open-webui:main
After the installation is completed, visit http://localhost:3000, select deepseek-r1:latest model and start using it.
V. Performance optimization and resource management
- Resource allocation: Select the appropriate model version according to the hardware configuration. Smaller models (such as 1.5B to 14B) perform well on standard hardware, while larger models (such as 32B and 70B) require more powerful GPU support.
- Memory management: Make sure the system has enough memory and storage space to avoid insufficient resources during runtime.
6. Frequently Asked Questions and Solutions
- Model download timeout: If there is a timeout problem when downloading the model, you can try to rerun the download command.
- Service startup failed: Make sure the Ollama service is installed and started correctly. If the service starts fail, you can try restarting the Ollama service.
7. Summary
Through the above steps, you can successfully deploy the DeepSeek model locally and interact with the model through Ollama or Open Web UI. Local deployment not only protects data privacy, but also flexibly adjusts model parameters according to requirements to meet the usage needs in different scenarios.
Attachment: DeepSeek usage skills and advanced strategies
(I) Optimize the first experience
Don't ignore some important options in your system settings when using DeepSeek for the first time. Be sure to check the "Optimization Mode" button, which is a key setting to improve the quality of AI output. It is like opening up the "Ren and Du meridians" of DeepSeek, which can make it perform better when generating content and answering questions.
(II) Rethinking prompt words
In the past, when using AI, we often rely on complex prompt word templates, but in the DeepSeek-R1 operating environment, this method may no longer be applicable. The new model is more sensitive to prompt words, and users should describe tasks directly and concisely as possible without relying on cumbersome examples. For example, if you want to write a New Year greeting to your elders in the Year of the Snake, and directly tell DeepSeek "write a New Year greeting to your elders in the Year of the Snake", it can automatically output texts of multiple styles for you to choose from, greatly simplifying the creative process and improving efficiency.
(III) Effective communication with AI
Expressing needs concisely: When you feel that DeepSeek's response is too complex and difficult to understand, you might as well try to re-express the needs in more concise and straightforward language. For example, you can use phrases such as "speaking human words" and "please explain in easy-to-understand language" to guide it to output more understandable content.
Multiple communication attempts: Communication with DeepSeek is a process of continuous adjustment. Don’t give up just because of an unsatisfactory answer. By trying many times, adjusting the way and content of the question, you will gradually find the best way to communicate with it and get more satisfactory answers. For example, when asking a complex technical question, if the first answer is not detailed enough, you can further ask for specific details and let it supplement and improve it.
(IV) Use stylized rewrite
DeepSeek allows users to specify styles to rewrite content, which provides more possibilities for creation. You can let it present a paragraph of text in Lu Xun's style and feel the unique literary charm; or let it imitate the expressions of well-known business writers and add professional color to commercial copywriting. For example, a normal product introduction copy is rewritten in the style of Jobs' speech at the product launch conference to make it more infectious and attractive.
(V) Create a custom knowledge base (advanced functions)
For users with specific needs, DeepSeek also supports uploading files to create a custom knowledge base. After uploading documents and materials related to your work and study, DeepSeek can provide you with more personalized and targeted answers and suggestions based on this knowledge. For example, corporate users can upload the company's internal rules and regulations, business materials, etc., so that DeepSeek becomes an intelligent assistant within the company; students can upload their own study notes, professional documents, etc. to help themselves learn and review better.
(VI) Multimodal interaction (partial functions)
Although some of DeepSeek's models are not completely multimodal, the company has launched models such as Janus Pro that have the ability to generate images and text comprehensions. In the function that supports multimodal interaction, users can not only enter text, but also upload pictures to ask questions or create. For example, upload a landscape photo and ask it to write a beautiful travel notation based on the photo; or ask a question related to the picture, such as "What is the style of the architecture in this picture?" It will give an accurate answer based on the content of the picture.
This is all about this article about DeepSeek local deployment. For more relevant DeepSeek local deployment content, please search for my previous articles or continue browsing the related articles below. I hope everyone will support me in the future!