Deploy DeepSeek Locally on Ubuntu: Build Your Private AI Assistant

Why Run a Large Model Locally?

Privacy: All code, documents, and conversations stay on your disk.

Zero latency: Response time depends only on your GPU, not network fluctuations.

Free: No token billing; you can call the model 24/7.

🛠️ Preparation: Check Your Hardware

Before starting, open a terminal and run nvidia-smi to verify your GPU.

Entry‑level (7B model): At least 8 GB VRAM (e.g., RTX 3060/4060) or 16 GB unified memory.

Advanced (33B/67B model): 24 GB+ VRAM (RTX 3090/4090 or multi‑GPU).

CPU‑only: Possible but expect high fan noise and slower inference.

Note: Ensure proprietary GPU drivers are installed. On Ubuntu you can use "Software & Updates → Additional Drivers".

🚀 Step 1: Install the AI Runtime – Ollama

Ollama is the de‑facto tool for running LLMs on Linux, comparable in simplicity to Docker. curl -fsSL https://ollama.com/install.sh | sh After installation, verify the service: systemctl status ollama If you see active (running), the engine is up.

📥 Step 2: Pull a DeepSeek Model

Ollama’s model library includes the DeepSeek‑R1 series, which excels at reasoning and code generation. Choose a model size that matches your VRAM:

7B (recommended for most users): Balanced speed and capability. ollama run deepseek-r1:7b 1.5B (lightweight): Fits older laptops. ollama run deepseek-r1:1.5b 32B/70B (performance beasts): For high‑end GPUs. ollama run deepseek-r1:32b Ollama will automatically download the model weights (several GB) and then drop you into an interactive prompt.

💬 Step 3: Hello, DeepSeek!

When the terminal shows the >>> prompt, you can start a conversation. For example:

>>> 请用 Python 写一个快速排序算法，并解释时间复杂度。

The model returns Python code and, using the <think> tag, displays its chain‑of‑thought reasoning, showcasing the R1 series’ strength. Exit with /bye.

🎨 Advanced: Add a Graphical UI with Open WebUI

If you prefer a ChatGPT‑like web interface or want to upload documents, run Open WebUI in Docker:

# Run Open WebUI container
docker run -d -p 3000:8080 \
  --add-host=host.docker.internal:host-gateway \
  -v open-webui:/app/backend/data \
  --name open-webui \
  --restart always \
  ghcr.io/open-webui/open-webui:main

After the container starts, open http://localhost:3000 in a browser, register an admin account (data is stored locally), and you will see the previously downloaded deepseek-r1:7b model ready for use.

🔮 Summary

Use the AI workstation to write code, debug logs, or translate documents.

Next article will cover Retrieval‑Augmented Generation (RAG) to feed Ubuntu documentation into DeepSeek, turning it into a true Ubuntu operations expert.