Deploy DeepSeek R1 Locally with Ollama: Step‑by‑Step Guide for Windows & Linux

Introduction

DeepSeek, a leading domestic inference large model, offers high performance and low cost, making it a focus in AI. Due to growing data privacy concerns, local deployment is preferred. This guide details how to deploy DeepSeek R1 with Ollama on Windows and Linux, including hardware requirements, deployment steps, and usage examples.

DeepSeek Model Overview

DeepSeek R1 series includes several versions for different scenarios and hardware:

Full‑scale DeepSeek‑R1 : 671B parameters, requires professional servers for high‑performance computing.

Distilled DeepSeek‑R1 : 1.5B‑70B parameters, lower VRAM demand, suitable for regular users and developers.

Quantized version : Uses model compression to further reduce VRAM, suitable for home PCs and mobile devices.

Hardware Requirements

Windows/Linux :

DeepSeek R1 VRAM Requirements

Standard (full‑precision) and quantized versions have different VRAM needs. Standard models range from ≥4 GB (1.5B) to ≥160 GB (671B). Quantized models reduce VRAM, e.g., 1.58‑bit version needs 160 GB, 2.51‑bit version needs 240 GB.

Deploy DeepSeek R1 with Ollama

Windows Deployment Steps

Install Ollama : Download the Windows installer from the Ollama website and install.

Download DeepSeek model :

Run the model :

Linux Deployment Steps

Install Ollama :

Download and run the model :

Performance optimization :

Visual Client Deployment Options

Open WebUI (recommended) : Docker‑based UI supporting multiple models, chat history, Markdown rendering, and API compatibility.

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

Access at http://localhost:3000.

Chatbox (lightweight client) : Cross‑platform tool; configure API type Ollama, address http://localhost:11434, and model name deepseek-r1:[version].

LobeChat (enterprise‑grade) : Docker deployment for team collaboration.

docker run -d -p 3210:3210 --name lobe-chat --restart always lobehub/lobe-chat

Set OLLAMA_BASE_URL to the local Ollama service.

Usage Examples

Code generation : Prompt “Write a Python product‑management API with JWT authentication and SQLAlchemy integration.” Returns full Flask API code with tests.

Financial data analysis : Prompt “Generate a CSI 300 trend analysis report with visualizations.” Returns a report with charts.

Creative content : Prompt “Write a cyber‑punk themed love poem.” Returns a stylized poem.

Deployment & Usage Recommendations

Hardware selection priority : VRAM > Memory > Storage. VRAM determines the maximum model size; memory should be at least 1.5× VRAM; SSD storage of 50 GB+ is advised.

Performance tuning tips :

Multi‑model management : Use ollama list to view local models and ollama cp to duplicate models for customization.

By following these methods, users can select appropriate model versions and interfaces based on hardware constraints, achieving a balance between performance and ease of use.

Conclusion

Local deployment of DeepSeek provides high‑performance AI services while protecting data privacy, delivering significant value for developers and enterprises alike. Give it a try and unlock the potential of on‑premise AI.