DeepSeek本地安装

✅ My System Information
OS: Ubuntu 22.04
CPU: Intel i5-12600KF
GPU: NVIDIA GTX 3060 (12GB VRAM)
CUDA Version: 12.1
cuDNN Version: 9.0.100
Python Version: 3.8.20 (Anaconda)
PyTorch Version: 12.1 (CUDA-enabled)

 

Since my Ubuntu 22.04 + RTX 3060 (12GB VRAM) + CUDA 12.1 + cuDNN 9.0.100 + PyTorch 12.1 setup is well-prepared, the best way to install DeepSeek locally is via Ollama.

This method is easy, optimized, and ensures proper GPU acceleration without manually handling model weights.

 

1. Open a terminal and run:

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

2. Restart your shell:

source ~/.bashrc

 3. Download DeepSeek Model

ollama pull deepseek-r1:1.5b

 

If I want a larger model and can manage VRAM usage:

ollama pull deepseek-r1

 

 

 4: Run DeepSeek Locally**
Start the DeepSeek model:

ollama run deepseek-r1:1.5b

 

For the larger model:

ollama run deepseek-r1