DeepSeek R1 Download and Installation Guide

Quick Start Guide

1. Download and Install Ollama

Windows Download macOS Download Linux Download

2. Install DeepSeek R1

ollama run deepseek-r1

Available Models

Model Name Size Installation Command Base Model
DeepSeek-R1-Distill-Qwen-1.5B 1.5B ollama run deepseek-r1:1.5b Qwen-2.5
DeepSeek-R1-Distill-Qwen-7B 7B ollama run deepseek-r1:7b Qwen-2.5
DeepSeek-R1-Distill-Llama-8B 8B ollama run deepseek-r1:8b Llama3.1
DeepSeek-R1-Distill-Qwen-14B 14B ollama run deepseek-r1:14b Qwen-2.5
DeepSeek-R1-Distill-Qwen-32B 32B ollama run deepseek-r1:32b Qwen-2.5
DeepSeek-R1-Distill-Llama-70B 70B ollama run deepseek-r1:70b Llama3.3

License Information

The model weights are licensed under the MIT License. DeepSeek-R1 series support commercial use, allow for any modifications and derivative works, including, but not limited to, distillation for training other LLMs.

Important License Notes:

  • The Qwen distilled models (1.5B, 7B, 14B, 32B) are derived from Qwen-2.5 series
    • Originally licensed under Apache 2.0 License
    • Finetuned with 800k samples curated with DeepSeek-R1
  • The Llama 8B distilled model
    • Derived from Llama3.1-8B-Base
    • Licensed under llama3.1 license
  • The Llama 70B distilled model
    • Derived from Llama3.3-70B-Instruct
    • Licensed under llama3.3 license

Model Performance Comparison

DeepSeek Model Performance Comparison

Benchmark Results

AIME 2024

DeepSeek-R1 achieves 79.8% accuracy, significantly outperforming OpenAI and other models

Codeforces

Outstanding 96.3% performance, leading in code generation capabilities

MATH-500

97.3% accuracy, demonstrating superior mathematical reasoning abilities

MMLU

90.8% accuracy in multi-task language understanding

Key Findings:

  • DeepSeek-R1 consistently outperforms OpenAI models across multiple benchmarks
  • Particularly strong in mathematical reasoning (MATH-500) and coding tasks (Codeforces)
  • DeepSeek-R1-32B shows balanced performance across different tasks
  • Superior performance in both academic (AIME) and practical (SWE-bench) applications

Key Features

Advanced Performance

State-of-the-art language model with exceptional coding and reasoning capabilities.

Local Execution

Run entirely on your machine for enhanced privacy and reduced latency.

Multi-language Support

Comprehensive support for major programming languages and frameworks.

Detailed Installation Guide

Windows Installation

  1. Download the Ollama Windows installer
  2. Run the installer as administrator
  3. Open Command Prompt or PowerShell
  4. Run: ollama run deepseek-r1

Usage Examples

Code Generation

            Query: "Write a Python function to calculate Fibonacci numbers"
            Response: ...

Performance Comparison

Mac Compatibility Guide

Apple Silicon Macs (M1/M2/M3)

Performance Comparison

Mac Model Recommended DeepSeek Version Performance
M3 Pro/Max/Ultra DeepSeek R1 7B Excellent - Full Speed
M2 Pro/Max DeepSeek R1 7B Very Good
M1 Pro/Max DeepSeek R1 7B Good
M1/M2/M3 DeepSeek R1 1.3B Good
Model Response Time Accuracy Memory Usage
DeepSeek R1 7B ~200ms 98% 16GB
DeepSeek R1 1.3B ~100ms 95% 8GB

Intel Macs

Mac Model Recommended DeepSeek Version Notes
MacBook Pro (2019-2021)
with dedicated GPU		 DeepSeek R1 7B Requires good cooling
Other Intel Macs DeepSeek R1 1.3B Limited performance

Important Notes

Model Response Time Accuracy Memory Usage

System Requirements

Model Version Required VRAM Recommended GPU
DeepSeek R1 1.3B 4GB VRAM NVIDIA GTX 1060 or better
DeepSeek R1 7B 8GB VRAM NVIDIA RTX 2060 or better

Troubleshooting

Insufficient VRAM

If you encounter VRAM issues, try using a smaller model variant or upgrade your GPU.

Download Errors

Check your internet connection and try again. If the issue persists, try using a VPN.

Need Help?

For technical support or questions, please contact us at [email protected]