Raspberry Pi AI In Hand With DeepSeek R1
The current LLMs are compact enough to run on a Raspberry Pi. This article demonstrates how to install and run the DeepSeek R1 model on a Raspberry Pi 5B. Let's start building a Raspberry Pi AI machine!
Software and Hardware Versions:
System: Raspbian OS Bookworm 64bit
Pi Model: Raspberry Pi 5B (8GRAM)
TF Card: 128G
Setup Ollama
Ollama allows us to easily run LLMs on various systems. To install Ollama on a Raspberry Pi, simply use the following command:
sudo apt install curl -y
curl -fsSL https://ollama.com/install.sh | sh
then either (or one at a time)
ollama run deepseek-r1:1.5b
Now you get a Raspberry Pi ai chatbot, you can chat with the DeepSeek R1 1.5B model on your Raspberry Pi. If you want to end the conversation, simply type /bye to exit the program.
Install Open-WebUI
If you want to have a ChatGPT-like experience, you can also install Open WebUI, turning your Raspberry Pi AI chatbot into a small Raspberry Pi AI chat server.
# create folder and python virtual environment
cd
mkdir webui
pip -m venv webui
source venv/bin/activate
# install open-webui
pip install open-webui
# start
open-webui serve
This will start the your Raspberry Pi AI server, which you can access at http://<raspberrypi_ip>:8080
Experience DeepSeek R1 on Raspberry Pi
Now you can start chatting with the LLM through the interface.
If you want to choose a different model, you can select it from the top left corner of the interface. If the model you want hasn’t been downloaded yet, you can simply click “Get from ollama.com” to download it. It‘s really convenient!
Here, we tested the DeepSeek R1 on Raspberry Pi models with 1.5B, and 7B parameters on Raspberry Pi 5B. The1.5B model taking about 1 minute and 20 seconds to finish the writing task. The 7B model runs significantly slower than the 1.5B model,taking about 4 minute and 15 seconds to finish the same writing task.
When the Raspberry Pi starts processing, the cooling fan gradually ramps up to full speed, and you can hear its whirring sound. For larger models, the inference time would clearly double or more, so we decided not to test further.
Raspberry Pi AI Server in Hand
All DeepSeek R1 models incorporate Chain of Thought (CoT) reasoning, making their responses more accurate and persuasive. LLMs are becoming increasingly intelligent while also reducing their hardware requirements. In the past, running an LLM might have required large server racks in a data center. Now, it can run on a device no bigger than the palm of your hand. We used a custom case and powered it with the PiSugar 3 Plus (use a 4000mAh battery to fit in the case), turning it into a portable LLM server!
Enjoy the power of AI!
Links:
Ollama: https://github.com/ollama/ollama
Open-Web UI: https://github.com/open-webui/open-webui
PiSugar server case: https://github.com/PiSugar/PiSugar-server-case/tree/main/pc_case