What is Distilled DeepSeek R1? How It Will Change Microsoft Copilot+ PCs

In an exciting new development for both mobile and desktop computing, Microsoft has expanded its AI portfolio by bringing the distilled versions of the powerful DeepSeek R1 models to Copilot+ PCs.

The DeepSeek R1 model, which has already made waves in the mobile world, is now ready for use on Windows platforms with Microsoft’s full support.

What is DeepSeek R1?

DeepSeek is an advanced AI model designed for deep learning, originally built with 671 billion parameters, making it one of the most sophisticated models in existence.

However, not every device can run such a large model due to memory and computational limitations. To address this, Microsoft has distilled the original DeepSeek R1 model into smaller, more manageable versions that can run on consumer-grade hardware.

The first distilled version, DeepSeek-R1-Distill-Qwen-1.5B, features just 1.5 billion parameters, offering a more compact yet highly capable AI solution.

Bringing DeepSeek to Copilot+ PCs

The distilled versions of DeepSeek will be available to devices powered by Snapdragon X chips, Intel Core Ultra 200V processors, and AMD Ryzen AI 9-based PCs.

These models have been specifically optimized for use on devices with Neural Processing Units (NPUs), ensuring they can perform efficiently without requiring specialized AI hardware.

Microsoft’s optimizations allow the DeepSeek model to deliver fast processing speeds, achieving a time to first token of just 130 milliseconds.

For short prompts under 64 tokens, it can process 16 tokens per second, ensuring rapid response times and impressive throughput for everyday users.

How Does Model Distillation Work?

Model distillation, also known as knowledge distillation, is the process of transferring the knowledge of a large, complex model into a smaller, more efficient one. While the smaller model may not match the full model in every aspect, it retains a large portion of its capabilities while being optimized to run on less powerful hardware.

In this case, the full DeepSeek R1 model is an impressive 671 billion parameters, but the distilled versions, such as the 1.5B model, are compact enough to run on consumer devices.

This approach has significant benefits, allowing developers to leverage the power of AI directly on personal computers without needing to rely on expensive dedicated AI hardware.

The models are available for download from Microsoft’s AI Toolkit, and users can even test them locally using tools like Visual Studio Code (VS Code) to explore their capabilities.

Microsoft’s Expanding AI Ecosystem

Microsoft has shown strong support for various AI models from a range of developers, including OpenAI (the creators of ChatGPT), Meta’s Llama, and Mistral.

With DeepSeek now entering the fold, Microsoft is further cementing its role as a major player in the AI space, offering developers access to a wide variety of models that cater to different use cases and hardware environments.

The versatility of the DeepSeek R1 distilled models ensures that even those working on lower-end PCs can take advantage of cutting-edge AI technology without sacrificing performance.

Whether you’re working on cloud-based applications or prefer a local solution, the new DeepSeek integration offers flexibility and power at your fingertips.

The addition of distilled DeepSeek R1 models to Copilot+ PCs is a significant move by Microsoft to bring powerful AI capabilities to everyday users.

By making these models available on consumer-grade hardware, Microsoft is lowering the barrier to entry for AI development, allowing a wider audience to benefit from this advanced technology.

Whether for developers or casual users, DeepSeek is poised to become an integral part of the AI landscape, thanks to its ability to run on standard personal devices.