Skip to content

Setup Guide: SmoLAgents with DeepSeek-R1 on Ollama

This guide will walk you through setting up SmoLAgents with a local DeepSeek-R1 model running on Ollama.

Prerequisites

  • Python 3.8+
  • Local GPU with minimum 12GB VRAM (recommended)
  • Basic understanding of Python and command-line tools
  • Windows, macOS, or Linux operating system

Step 1: Install Ollama

  1. Visit ollama.ai to download the appropriate version for your operating system
  2. Install Ollama following the instructions for your platform
  3. After installation, open a terminal or command prompt and verify Ollama is working:
1
ollama --version

Step 2: Download DeepSeek-R1 Model

  1. Pull the DeepSeek-R1 7B model (this will take some time depending on your internet connection):
1
ollama pull deepseek-r1:7b
  1. Verify the model was downloaded successfully:
1
ollama list

You should see deepseek-r1:7b in the list of available models.

Step 3: Set Up Python Environment

  1. Create a virtual environment (recommended):
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
# Create directory for project
mkdir smolagents-tutorial
cd smolagents-tutorial

# Create virtual environment
python -m venv venv

# Activate virtual environment
# On Windows:
venv\Scripts\activate
# On macOS/Linux:
source venv/bin/activate

  1. Create a requirements.txt file with the content provided in this tutorial or download it from the repository

  2. Install the required packages:

1
pip install -r requirements.txt

Step 4: Test Basic Functionality

  1. Create a simple test script test_smolagents.py using the code provided in the tutorial

  2. Run the test script to verify SmoLAgents is installed correctly:

1
python test_smolagents.py
  1. Check that Ollama is running the server:
1
2
# Start Ollama server if not already running
ollama serve

Step 5: Run the Example Agents

  1. Start with the simple agent:
1
python simple_agent.py
  1. Try more advanced examples like the code agent and data analysis agent:
1
2
python code_agent.py
python data_analysis_agent.py
  1. For the knowledge graph example, ensure you have NetworkX installed:
1
2
pip install networkx matplotlib
python knowledge_graph_agent.py

Troubleshooting

Model Loading Issues

If you encounter issues loading the model:

1
2
# Restart the Ollama server
ollama serve

Out of Memory Errors

If you see CUDA out of memory errors:

  1. Close other applications using GPU memory
  2. Try a smaller model like deepseek-r1:6.7b-dpo-q4_0 which uses less memory

Connection Issues

If the agent can't connect to Ollama:

  1. Verify Ollama is running: ollama serve
  2. Check the API endpoint in your code: should be http://localhost:11434/v1
  3. Ensure no firewall is blocking the connection

Next Steps

After successfully running the examples, try:

  1. Creating your own custom tools
  2. Experimenting with different agent types (JSONAgent, ReActAgent)
  3. Integrating with other data sources or APIs
  4. Building a complete application using the agent framework

Happy coding with SmoLAgents and DeepSeek-R1!