How to Deploy Large Language Models (LLMs) with Ollama

Ollama is an open-source platform for running large language models (LLMs) locally. Ollama supports most open-source Large Language models (LLMs) including Llama 3, DeepSeek R1, Mistral, Phi-4, and Gemma 2 models, you can run locally without an internet connection. It enhances security, privacy, and provides greater control over a model's performance, allowing you to customize it on your workstation.

Ollama includes a default models repository which lets you search and download LLMs to run on your workstation. It also supports OpenWebUI as a graphical web interface for running models without using the command line interface (CLI). It works on Linux, Windows, and macOS without any cloud-based APIs requirements to run models.

This article explains how to install Ollama and run large language models (LLMs) with all required dependencies on your workstation.

Download and Install Ollama

Ollama is compatible with Linux, macOS, and Windows. You can download and install Ollama using the official release package or script from the official website. Follow the steps below to download and install the latest Ollama version on your Linux, macOS, or Windows workstation.

Download LLMs Using Ollama

Ollama downloads models using the ollama pull command that fetches the model to run on your system. You can download models from the Ollama models repository. Follow the steps below to download LLMs using Ollama to run on your workstation.

1. Run the Ollama with the pull option to download a model from the official repository.

   console Copy

   $ ollama pull [model]
   

   Explain Code

2. For example, download Mistral using Ollama.

   console Copy

   $ ollama pull mistral
   

   Explain Code

3. Download the Deepseek-R1-Distrill-Qwen model with 1.5B parameters using Ollama.

   console Copy

   $ ollama pull deepseek-r1:1.5b
   

   Explain Code

4. Download Llama 3.3 using Ollama.

   console Copy

   $ ollama pull llama3.3
   

   Explain Code

   The Llama 3.3 model is 40GB in size. Verify that you have enough storage space to run this model before downloading it.

5. List all locally available models using Ollama.

   console Copy

   $ ollama list
   

   Explain Code

   Your output should be similar to the one below.

   NAME                ID              SIZE      MODIFIED
   llama3.3:latest     a6eb4748fd29    42 GB     21 seconds ago
   deepseek-r1:1.5b    a42b25d8c10a    1.1 GB    4 minutes ago
   mistral:latest      f974a74358d6    4.1 GB    26 minutes ago

Use Ollama to Run Models

You can run, pull and initialize large language models directly from the Ollama repository or locally from the available models using the ollama run command. Check the required system resources for each AI model before running it using Ollama. Follow the steps below to run models using Ollama and test the performance of each model on your workstation.

1. List all locally available models.

   console Copy

   $ ollama list
   

   Explain Code

2. Run a model directly using Ollama. For example, run the qwen 2.5 instruct model with 1.5B parameters.

   console Copy

   $ ollama run qwen2.5:1.5b
   

   Explain Code

3. Enter a new prompt such as Give me two lines of text about Vultr.

   >>> Send a message (/? for help)

4. Monitor the model output and results in your terminal.

   >>> Give me two lines of text about Vultr
   Vultr is a popular cloud hosting provider known for its affordable pricing and user-friendly interface, making it an
   ideal choice for both small businesses and individual users looking to host websites or run servers. It offers various
   storage options starting at just $5/month, with plans that cater to different needs ranging from basic to advanced
   configurations.

5. Enter /bye to exit Ollama.

   >>> /bye

6. Run a model that's available on your workstation. For example, run the Deepseek R1 model you downloaded earlier.

   console Copy

   $ ollama run deepseek-r1:1.5b
   

   Explain Code

7. Enter a prompt such as Generate a recursive fractal pattern description using only mathematical notations and symbolic logic.

   >>> Send a message (/? for help)

8. Monitor the model output and results in your terminal.

   <think>
   Okay, so I need to create a recursive fractal pattern using only mathematical notation and symbols. Hmm, that's an
   interesting challenge. Let me think about what I know about fractals and how they can be represented mathematically.
   ...................................

9. Enter /bye to exit Ollama.

   >>> /bye

   Different models excel at specific tasks and running them using Ollama allows you to test the performance and model capabilities. Use the ollama run command to run locally available models, or pull the latest version from the Ollama models repository to run on your system.

Manage Available Models Using Ollama

Downloading multiple LLMs on your system is a complex task that requires continuous management using Ollama. Use the ollama list command to view all available and delete any versions you no longer need on your workstation. Follow the steps below to view and manage all available models using Ollama.

1. List all locally available models.

   console Copy

   $ ollama list
   

   Explain Code

2. Show information about a specific model. For example, display the Llama 3.3 model information.

   console Copy

   $ ollama show llama3.3
   

   Explain Code

   Output:

    Model
      architecture        llama
      parameters          70.6B
      context length      131072
      embedding length    8192
      quantization        Q4_K_M
   
    Parameters
      stop    "<|start_header_id|>"
      stop    "<|end_header_id|>"
      stop    "<|eot_id|>"
   
    License
      LLAMA 3.3 COMMUNITY LICENSE AGREEMENT
      Llama 3.3 Version Release Date: December 6, 2024

3. Stop a model that's actively running using Ollama.

   console Copy

   $ ollama stop [model-name]
   

   Explain Code

4. For example, stop the Deepseek R1 model.

   console Copy

   $ ollama stop deepseek-r1:1.5b
   

   Explain Code

5. Remove a model from your workstation. For example, remove the mistral model you pulled earlier.

   console Copy

   $ ollama rm mistral
   

   Explain Code

   Output:

   deleted 'mistral'

6. List all locally available models and verify that the model is unavailable.

   console Copy

   $ ollama list
   

   Explain Code

Set Ollama Environment Variables

Ollama variables modify the application behavior and allow you to effectively run models on your workstation. The following are the most used Ollama variables.

• OLLAMA_HOST: Specifies the Ollama server address.
• OLLAMA_GPU_OVERHEAD: Reserve a part of VRAM per GPU in bytes.
• OLLAMA_MODELS: Specifies a custom directory to store the model files on your system.
• OLLAMA_KEEP_ALIVE: Controls how long models can stay in memory.
• OLLAMA_DEBUG: Enables additional debug information while running LLMs.
• OLLAMA_FLASH_ATTENTION: Activates experimental optimizations for attention mechanisms
• OLLAMA_NOHISTORY: Disables readline history while running LLMs.
• OLLAMA_NOPRUNE: Disables pruning of model blobs at system boot or startup.
• OLLAMA_ORIGINS: Configures origin URLs with access to the Ollama server.

Follow the steps below to set Ollama environment variables based on your operating system.

Set Ollama Variables on Linux

1. Open the Ollama system service file.

   console Copy

   $ sudo vim /etc/systemd/system/ollama.service
   

   Explain Code

2. Add the Environment command after the [service] declaration and include an Ollama variable. For example, enter Environment="OLLAMA_DEBUG=1" to enable debugging.

   [Service]
   Environment="OLLAMA_DEBUG=1"

   • Add another Ollama environment variable such as OLLAMA_HOST=0.0.0.0:11434 to allow other hosts to connect to the Ollama server.

     ini Copy

     [Service]
     Environment="OLLAMA_DEBUG=1"
     Environment="OLLAMA_HOST=0.0.0.0:11434"
     

     Explain Code

     Save and close the file.

     Setting the Ollama host address to 0.0.0.0 allows other hosts to connect and access Ollama using the specified port.

3. Reload systemd to apply the service configuration changes.

   console Copy

   $ sudo systemctl daemon-reload
   

   Explain Code

4. Restart Ollama to apply the environment variables.

   console Copy

   $ sudo systemctl restart ollama
   

   Explain Code

Set Ollama Variables on macOS

1. Quit Ollama if it's running in your terminal session.

2. Use the launchctl command to set environment variables on Mac. For example, set the OLLAMA_HOST variable.

   console Copy

   $ launchctl setenv OLLAMA_HOST "0.0.0.0"
   

   Explain Code

3. Start Ollama to apply the variable changes.

   console Copy

   $ ollama serve
   

   Explain Code

Set Ollama Variables on Windows

1. Quit Ollama if it's running.

2. Open the Windows search menu, search for Environment Variables, and select Edit the System Variables from the search results.

3. Click Environment Variables in the open System Properties window.

   

4. Click New to create a new variable.

5. Enter your Ollama variable in Variable name.

6. Enter a value in the Variable value.

   

7. Click OK to save the environment variable.

8. Click OK to close the Environment Variables prompt.

9. Click Apply to apply the variable changes and close the System Properties dialog.

Conclusion

You have installed Ollama and run large language models (LLMs) on your workstation. Ollama is an efficient platform for running LLMs locally or on a remote server. Use Ollama variables to create a dedicated workstation to run and access your models remotely from any device, including mobile devices. For more information and model examples, visit the Ollama GitHub repository.