Experimenting with DeepSeek R1 Locally

After reading a few posts, in particular Simon Willison’s DeepSeek-R1 and exploring DeepSeek-R1-Distill-Llama-8B, I took the plunge to install and run DeepSeek R1 on my local machine.

Following instructions from Simon’s post and How to Use DeepSeek R1 in Visual Studio Code with Cline, I managed to get everything running this way on my terminal:

  1. Install Ollama:

Download it from ollama.com, or install via homebrew like I did, to manage local AI models.

brew install ollama

  1. Run Ollama:

ollama serve

  1. Download the model

I went for the 14B model, that seems to be a good enough model to run locally. But there are other options: 1.5B, 8B, 14B, 32B, and 70B. Pick what’s best for your use case.

  • 1.5B: The smallest and lightest, great for testing or machines with limited resources.
  • 8B: A good balance of performance and resource usage, a solid starting point for most users.
  • 14B: Stepping up in capability, offering improved performance for more demanding tasks.
  • 32B: Getting into the larger, more powerful models, requiring more GPU power.
  • 70B: The largest and most intelligent, designed for top-tier performance but needs significant GPU resources.

ollama pull deepseek-r1:14b

  1. Install LLM Ollama plugin

Using Simon’s approach and open source tool, LLM, that I like.

llm install llm-ollama

  1. Run the LLM chat interface

llm chat -m 'deepseek-r1:14b'

  1. Profit!

That’s it. DeepSeek R1 14B running locally in chat mode!