DeepSeek-R1-0528 is a cutting-edge open-source large language model (LLM) designed for developers, researchers, and AI enthusiasts. With state-of-the-art benchmark performance, advanced reasoning capabilities and support for JSON output and function calling, this model stands out for both experimentation and production use.

In this guide, you'll learn how to run and install DeepSeek-R1-0528 on your local machine using Ollama, vLLM, and Hugging Face Transformers.

Overview of DeepSeek-R1-0528

DeepSeek-R1-0528 is the latest entry in the DeepSeek-R1 series and offers:

• Enhanced reasoning and factual accuracy
• Reduced hallucinations for better reliability
• JSON output support and function calling
• Commercial-use friendly MIT License

The model is freely available with open-source weights on Hugging Face.

Hardware Requirements

Before setting up DeepSeek-R1-0528, ensure your hardware meets the minimum system requirements based on the model size.

Tip: Use quantized versions for better performance on consumer-grade GPUs.

Installation Methods

You can install and run DeepSeek-R1-0528 locally using three main methods:

• Ollama – Fastest and easiest setup
• vLLM – High performance for production environments
• Transformers – Full flexibility for development pipelines

Method 1: Installing DeepSeek-R1-0528 with Ollama

Ollama simplifies LLM deployment and is ideal for getting started quickly.

Step-by-Step Setup

1. Check GPU Compatibility

nvidia-smi

2. Update System and Install Dependencies

sudo apt-get update
sudo apt-get install pciutils -y

3. Install Ollama

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

4. Start the Ollama Server

ollama serve

5. Verify Installation

ollama

6. Install the DeepSeek-R1-0528 Model

Currently available model (quantized):

ollama run hf.co/unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUF:Q4_K_XL

7. Run Inference

Use the terminal interface to send prompts and receive real-time responses.

Method 2: Installing DeepSeek-R1-0528 with vLLM

vLLM is ideal for high-throughput inference in scalable environments.

Requirements

• Python 3.8+
• CUDA-enabled GPU
• pip

Installation Steps

1. Install vLLM

pip install vllm

2. Download Model Weights

From Hugging Face:

• Visit: https://huggingface.co/deepseek-ai/DeepSeek-R1-0528

3. Launch the API Server

python -m vllm.entrypoints.openai.api_server \
  --model deepseek-ai/DeepSeek-R1-0528 \
  --tokenizer deepseek-ai/DeepSeek-R1-0528

4. Query the Model

Use the OpenAI-compatible API to send and receive prompts.

Method 3: Installing DeepSeek-R1-0528 with Transformers

Use this method for programmatic integration and full control over inference.

Requirements

• Python 3.8+
• PyTorch
• Transformers

Installation Steps

1. Install Required Packages

pip install torch transformers

2. Load the Model in Python

from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("deepseek-ai/DeepSeek-R1-0528")
model = AutoModelForCausalLM.from_pretrained("deepseek-ai/DeepSeek-R1-0528")

prompt = "Explain the difference between monorepos and turborepos."
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Prompt Engineering and Usage

DeepSeek-R1-0528 supports structured prompts, file handling, and web integration.

System Prompt Example

The assistant is DeepSeek-R1, created by DeepSeek。
Today is Monday, May 28, 2025.

Temperature Setting

• Default: 0.6

File Upload Template

file_template = """[file name]: {file_name}
[file content begin]
{file_content}
[file content end]
{question}"""

Web Search Template (English)

search_answer_en_template = '''
# The following contents are the search results related to the user's message:
{search_results}
...
# The user's message is:
{question}'''

Model Features and Improvements

• Superior Benchmarks – Outperforms previous releases on multiple tasks
• Reduced Hallucinations – More reliable for critical use cases
• Function Calling – Easy tool and API integration
• MIT License – Open for commercial use and distillation

Troubleshooting and Optimization

Common Issues

• Out of Memory – Use quantized models or reduce batch size
• Slow Inference – Optimize SSD and GPU usage
• Dependency Errors – Ensure Python, CUDA, and driver compatibility

Optimization Tips

• Use Ollama for quick experimentation
• Deploy vLLM for scalable APIs
• Choose quantized models for consumer-grade GPUs

Security and Best Practices

• Virtual Environments: Use venv or conda for dependency isolation
• Data Handling: Ensure compliance with data privacy laws
• Model Maintenance: Regularly update libraries and model versions

License and Commercial Use

DeepSeek-R1-0528 is licensed under the MIT License, allowing unrestricted commercial use, modification, and redistribution.

Conclusion

DeepSeek-R1-0528 empowers you to run a powerful, open-source LLM locally with a method that fits your workflow—Ollama for simplicity, vLLM for scalability, or Transformers for flexibility.

References

1. Run DeepSeek Janus-Pro 7B on Mac: A Comprehensive Guide Using ComfyUI
2. Run DeepSeek Janus-Pro 7B on Mac: Step-by-Step Guide
3. Run DeepSeek Janus-Pro 7B on Windows: A Complete Installation Guide
4. DeepSeek R1 0528 vs Google Gemini 2.5 Pro