Running and Serving LLMs with LM Studio

Overview

LM Studio is a powerful GUI-based wrapper for llama.cpp and also provides an OpenAI compliant endpoint for local model serving. LM Studio provides a simple but powerful interface to easily download and deploy models. LM Studio offers both Vulkan and AMD ROCm™ software backends (called runtimes) for AMD users.

What You’ll Learn

• How to configure and use LM Studio to leverage your local hardware
• Test and manage LLMs in a completely offline environment
• Serve models via OpenAI Compatible API to power custom workflows and apps

if command -v rocminfo >/dev/null 2>&1 && rocminfo | grep -q gfx; then
  echo "OK: GPU visible to ROCm"
else
  echo "GPU not visible via rocminfo" >&2
  exit 1
fi

sudo apt install pipx
pipx ensurepath
pipx install amd-debug-tools

amd-ttm

amd-ttm --set <NUM>

Installing Software Prerequisites

LM Studio

Downloading Models

Chatting with an LLM

Learn how to start chatting with a ChatGPT-grade LLM completely locally.

1. Open LMStudio.
2. Press Ctrl + L to open the Model Loader, select Manually choose model load parameters, and click on ${model_name}
3. Make sure “show advanced settings” is checked.
4. Change Context Length as desired. Higher context length means more model memory, but more system memory used. Recommended for this playbook is 4096.
5. Make sure GPU Offload is set to maximum and Flash Attention is On (Cache Quantizations can remain off)
6. Check Remember settings and click on Load Model.
7. If not in the chat window, press Ctrl + 1 or click on the 👾 button on the top left of the screen.
8. Send a message and start interacting with the model!

Serve LLMs through an OpenAI compatible endpoint

LM Studio also offers an OpenAI compliant endpoint in the form of LM Studio Server. This has already been demonstrated in an agentic coding workflow with Cline here. Another common use case is connecting LM Studio Server to any web application (React, Node.js, Python) by sending standard HTTP requests to the inference endpoint.

To set up LM Studio Server, use the following instructions:

1. On the left hand side, click on the Developer tab (command line icon) or Ctrl + 2 and then click on Server Settings.
2. (Optional): If you want to serve the model over your LAN, check Serve on Local Network. If you want to use with a website or extensive calling within VS Code, check Enable CORS.
3. On the upper left corner, make sure the server is running by clicking on the toggle button in front of Status.
4. An OpenAI compliant endpoint will now be running. The address is typically at http://127.0.0.1:1234
5. If a model is not already loaded, you can load it by clicking Load Model and following the previously mentioned steps.

This model will now be accessible through the LM Studio Server endpoint and will support OpenAI endpoints including:

Example: Pinging your Endpoint

Having just created the OpenAI Compatible endpoint, let’s look at how to integrate this into a Python developer environment (such as VSCode) and use your system as a local API Provider.

1. Create a Python virtual environment:

sudo apt update
sudo apt install -y python3-venv
python3 -m venv lmstudio-env --system-site-packages
source lmstudio-env/bin/activate

sudo usermod -aG render,video $LOGNAME

On Linux, open a terminal in the directory of your choice and follow the commands to create a venv.

sudo apt update
sudo apt install -y python3-venv
python3 -m venv lmstudio-env
source lmstudio-env/bin/activate

python -m venv lmstudio-env --system-site-packages
lmstudio-env\Scripts\activate

python -m venv lmstudio-env
lmstudio-env\Scripts\activate

2. Install the OpenAI package

pip install openai

3. Run the following script to ping the endpoint we have just created.

from openai import OpenAI

# Initialize the client specifically for your local server
# The API key is required by the library but ignored by LM Studio
client = OpenAI(
    base_url="http://localhost:1234/v1",
    api_key="lm-studio"
)
print("Attempting to connect to local LM Studio server...")

try:
    # Create a simple chat completion request
    completion = client.chat.completions.create(
        model="local-model", # The model identifier is optional in local mode
        messages=[
            {"role": "system", "content": "You are a helpful coding assistant."},
            {"role": "user", "content": "Explain Python decorators in 1 sentence"}
        ],
        temperature=0.7,
    )
    # Print the response
    print("\nConnection Successful! Server Response:\n")
    print(completion.choices[0].message.content)

except Exception as e:
    print(f"\nConnection Failed: {e}. Ensure LM Studio server is running on port 1234.")

(Optional): Swapping between Runtimes

1. Press Ctrl + Shift + R on your keyboard. Alternatively click on the Discover tab (Magnifying Glass) on the left-hand side and then click on Runtime in the pop up.
2. You should then see Runtime Selections, where the dropdown menu can be used to change the runtime.

Next Steps

• Custom App Integration: Integrate your own Python scripts or applications using the local OpenAI-compatible API.
• Advanced Frontends: Connect powerful interfaces like Open WebUI to your server for chat history and persona management.

For more documentation, please visit: https://lmstudio.ai/docs/developer