**DeepSeek, created by China DeepSeek, has set off a "Made in China Intelligent" storm around the world! This revolutionary app not only topped the App Store free charts in both China and the United States, but also swept the global market with an astonishing record of four times more downloads in a single day than ChatGPT. With only US$5 million in R&D investment, it has completely surpassed the multi-billion-dollar AI projects of Silicon Valley giants such as OpenAI and Google, and used its hard-core technical strength to interpret the efficiency of "Chinese-style innovation"! This small-scale technological counterattack has not only refreshed the global AI industry landscape, but also allowed the world to witness the disruptive power of Eastern wisdom. **

**This article will teach you how to deploy DeepSeek locally so that you can use DeepSeek anytime and anywhere. **

liunx deployment

Ollama is currently the simplest local model running framework that automatically handles dependencies and GPU acceleration.

1. Install Ollama

Open a terminal and execute the following script:

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

2. Start the service and run DeepSeek

Select the model version according to your video memory size (take DeepSeek-R1 as an example):

  • Version 1.5b (available with integrated graphics/core graphics):ollama run deepseek-r1:1.5b
  • 7b version (at least 8G video memory):ollama run deepseek-r1:7b
  • 14b version (at least 12G video memory):ollama run deepseek-r1:14b

After entering the command, the system will automatically download the model and enter the conversation mode.

3. API call

Ollama defaults to11434The port opens the API service, which you can call through Python:

import requests

json_data = {
    "model": "deepseek-r1:7b",
    "prompt": "为什么天空是蓝色的？",
    "stream": False
}
response = requests.post("http://localhost:11434/api/generate", json=json_data)
print(response.json()['response'])

If you need extremely high throughput, or need to use it in a production environment, vLLM is a better choice.

1. Prepare the environment

It is recommended to use Conda to create a virtual environment and install PyTorch with CUDA support:

conda create -n deepseek python=3.10 -y
conda activate deepseek
pip install vllm

2. Download model

Download the weights file for DeepSeek-R1 from Hugging Face or ModelScope.

3. Start the OpenAI compatible server

python -m vllm.entrypoints.openai.api_server \
    --model /path/to/deepseek-r1-weights \
    --tensor-parallel-size 1 \ # 如果有多个GPU，修改此数值
    --gpu-memory-utilization 0.9 \
    --port 8000

Option 3: Deploy using Docker (environment isolation)

If you prefer to use a containerized solution, you can start DeepSeek with the web interface (Open WebUI) using the following command:

docker run -d -p 3000:8080 --add-host=host.docker.internal:host-gateway -v open-webui:/app/backend/data --name open-webui ghcr.io/open-webui/open-webui:main

*Note: This solution requires you to first install Ollama according to solution 1. *

Troubleshooting

  1. Out of Video Memory (OOM): If startup fails, please try to use a version with smaller parameters (such as 1.5b or 7b), or turn on the quantized version (GGUF/AWQ).
  2. Driver issue: Please make sure the NVIDIA driver is installed. implementnvidia-smiCheck whether the GPU is recognized normally.
  3. Permission issue: IfcurlThe installation failed, please add before the commandsudo。

window deployment

1.Ollama download and install

Ollama is a lightweight local AI model running framework that can run various open source large language models locally (such as Llama, Mistral, etc.)

Browser input URL: https://ollama.com/

Select a system to download

After installation, Ollama is already running. It is a CMD command tool. We can enter ollama on the command line to verify whether the installation is successful.

If the content in the picture below appears, it means the download is successful.

Install the DeepSeek-r1 model Still on the Ollama website just now, select the Model module and select the deepseek-r1 model

There are many versions found in the search, the difference is that the parameters are different.

1.5b，7b，8b，14b，32b，70b，671b

The memory size required for each version is different.

If your computer has 8G of running memory, you can download 1.5b, 7b, and 8b distilled models.

If your computer has 16G of running memory, you can download the 14b distilled model.

I choose the 14b model here. The larger the parameters, the better the effect of using DeepSeek.

Use ollama run deepseek-r1:14b to download

Here I failed the first time, so I tried twice before I succeeded.

We can also enter ollama list in the command line to check whether the model is successfully downloaded. The content in the figure below indicates that the download is successful.

Enter ollama run deepseek-r1:14b to run the model

After successful startup, we can enter the questions we want to ask. The model will first perform in-depth thinking (that is, where the think tag is included), and after the thinking is completed, the results of our questions will be fed back.

Note: Since ollama downloads the model to the C drive by default, if your C drive space is as stretched as mine

Then we need to change the download location of the model

Edit environment variables

Create a new system environment variable

Please be sure to write OLLAMA_MODELS in the variable name

Then the variable value is the path to which you downloaded the model

Then you also need to go to the user environment variables and add two new environment variables

OLLAMA_HOST：0.0.0.0

OLLAMA_ORIGINS：*

After setting everything, save it and restart ollama

2.chatBox download and install

Chatbox AI官网：办公学习的AI好助手，全平台AI客户端，官方免费下载

After we download the model through ollama, we can use deepseek on the command line, but the form of the command line is still a bit unsightly, so we can use chatBox, which has a beautiful UI and can be used as long as it is connected to ollama's API.

It can also be downloaded on multiple platforms. I chose windows download here.

After the installation starts, click Settings

Model provider chooses ollama API

Then now you can select the model

can be used successfully

**Note: When choosing the size of the distillation model, you need to choose it based on the actual situation of your computer. This will affect the speed and effect of the model's answer. **