When installing Unsloth Studio on DGX Spark systems, the default installation process may fail or lead to ModuleNotFoundError during startup. This is primarily due to the ARM64 (aarch64) architecture of the DGX Spark, which can cause certain Python dependencies to be missed or incorrectly configured during the automated setup.
Symptoms:
When running unsloth studio -H 0.0.0.0 -p 8888, the application crashes with errors such as:
ModuleNotFoundError: No module named 'structlog'
ModuleNotFoundError: No module named 'fastapi'
ModuleNotFoundError: No module named 'diceware'
To fix this, you must manually install the missing dependencies into the active Unsloth environment before launching the studio.
Based on the system logs, the ARM-based environment requires explicit installation of three key packages that may not have registered during the initial script run.
Run the following commands in your terminal (ensure you are in the correct virtual environment):
pip install structlog
pip install diceware
# Install web framework dependency
pip install fastapi
Once the dependencies are installed, start the Studio using the host and port flags to ensure it is accessible across your network:
unsloth studio -H 0.0.0.0 -p 8888
Upon successful manual installation, the terminal should display the following indicators:
Hardware Detection: Hardware detected: CUDA -- NVIDIA GB10 (or your specific DGX Spark GPU).
Admin Credentials: A generated username (unsloth) and a bootstrap password path.
Status: INFO: Application startup complete and Uvicorn running on [http://0.0.0.0:8888](http://0.0.0.0:8888).
Result:
You should now be able to access the Unsloth Studio UI via your browser at http://<your-dgx-ip>:8888 and begin chatting with or training models.
Note: If you encounter further
ModuleNotFoundErrortraces, simplypip installthe specific module named in the last line of the traceback and restart the studio.