This post provides a way of setting up a prompting environment to run DeepSeek-R1 on various computing resources, be it a computer or server node or cluster. DeepSeek-R1 has demonstrated competitive performance and has attracted attention recently. Given that DeepSeek-R1 is publicly available, this post explains how to set it up and use it on computers or clusters.

The system requirements to run different versions of DeepSeek-R1 vary, yet that’s another topic, since the focus of the post is on installation and configuration. For a discussion about system requirements, see for example this post.

The present post focuses on installing the involved software on Linux from the command line without root access. The reason for this choise is to ensure that the setup is available both on Linux computers and on computer clusters, where typically root access is not available.

Ollama

In this post, Ollama is used for downloading and running DeepSeek-R1. First, the Ollama source code is downloaded under ~/opt/ollama, and then a soft link to the ollama binary is created in ~/bin. Assuming that ~/bin is in the user’s PATH environment variable, it suffices to source ~/.bashrc for the ollama binary to be accessible from the command line.

mkdir -p ~/opt/ollama
curl -L https://ollama.com/download/ollama-linux-amd64.tgz -o ollama-linux-amd64.tgz
tar -C ~/opt/ollama -xzf ollama-linux-amd64.tgz

mkdir ~/bin
ln -s ~/opt/ollama/bin/ollama ~/bin

source ~/.bashrc 

Typically, to run Ollama, its server can be started as follows:

ollama serve

However, it is possible that the default port is already occupied, in which case the Ollama server fails to start; I have experienced this issue on a remote computer node. This GitHub issue explains how to circumvent the problem by changing the port via the OLLAMA_HOST environment variable. Indeed, setting the port to 11435, instead of the default 11434, as shown below, solved the issue for me:

OLLAMA_HOST=127.0.0.1:11435 ollama serve

Having started the Ollama server, it can be checked whether it runs by starting a new terminal and running the following command, which prints the Ollama version:

ollama -v

If the Ollama server runs on a computer node or cluster, rather than a personal computer, then a second terminal can be started by running the screen shell command or by pressing F2 if tmux is used for terminal management. If tmux is used, then byobu-enable needs to be run first to turn on tmux.

At the moment of writing, there is no direct control over the choice of GPU or CPU nodes. Ollama uses all the GPU resources it autodetects; see this GitHub issue.

DeepSeek from the shell

Subject to the available system resources a version of DeepSeek-R1 is chosen. For demonstration purposes, the smallest deepseek-r1:1.5b version is pulled and run using the command below:

ollama run deepseek-r1:1.5b

Alternative quantized, distilled or full versions of DeepSeek-R1 include deepseek-r1:8b, deepseek-r1:14b, deepseek-r1:32b and deepseek-r1:70b.

The last command starts DeepSeek-R1 prompt. To terminate the prompt, the user can either type /bye on the prompt or press Ctrl+d.

DeepSeek from the Chatbox

Instead of interacting with DeepSeek-R1 from the shell, it is possible to do so from Chatbox. To install Chatbox from the command line, the following commands can be run:

mkdir -p ~/opt/chatbox
wget -P ~/opt/chatbox https://github.com/Bin-Huang/chatbox/releases/download/v0.10.4/Chatbox.CE-0.10.4-x86_64.AppImage
chmod u+x ~/opt/chatbox/Chatbox.CE-0.10.4-x86_64.AppImage

echo '\#!/bin/bash' > ~/opt/chatbox/Chatbox.CE-0.10.4-x86_64.sh
echo '~/opt/chatbox/Chatbox.CE-0.10.4-x86_64.AppImage --no-sandbox' >> ~/opt/chatbox/Chatbox.CE-0.10.4-x86_64.sh
chmod u+x ~/opt/chatbox/Chatbox.CE-0.10.4-x86_64.sh

ln -s ~/opt/chatbox/Chatbox.CE-0.10.4-x86_64.sh ~/bin/chatbox

source ~/.bashrc

Having executed the above commands, Chatbox can be started by typing chatbox on the command line. It is noted that the Bash script ~/opt/chatbox/Chatbox.CE-0.10.4-x86_64.sh was set up in order to be able to call Chatbox with the --no-sandbox option, thus avoiding the error message below, which I encountered:

The SUID sandbox helper binary was found, but is not configured correctly

After starting Chatbox, the model can be configured by clicking on Settings and by setting the following entries:

• Model Provider: OLLAMA.
• API Host: http://127.0.0.1:11434, http://127.0.0.1:11435 or any other combination of localhost and port used to start Ollama.
• Model: deepseek-r1:1.5b or any other pulled model present in the dropdown menu.

Apart from Ollama and Chatbox, the DeepSeek API can be integrated with other popular software. See this GitHub repository for alternative DeepSeek integrations.