I'm trying to use SLEAP (Social LEAP Estimates Animal Poses) to analyze animal movement data, but I’m running into issues with enabling GPU acceleration in Google Colab.

Since SLEAP only supports NVIDIA GPUs—and my personal machine has an AMD GPU—I’m attempting to run everything in Colab instead. I’ve already switched the runtime type to GPU (T4), but it seems like SLEAP is not detecting or using the GPU.

To replicate my local working environment (which uses Python 3.7), I used the following setup in a Colab notebook:

# Install Python 3.7 and create virtual environment
!apt-get install -y python3.7 python3.7-venv
!python3.7 -m venv /sleapvenv

# Fix for python-rapidjson build issue
!/sleapvenv/bin/pip install python-rapidjson==1.9

# Activate venv and install SLEAP
!/sleapvenv/bin/pip uninstall -qqq -y opencv-python opencv-contrib-python
!/sleapvenv/bin/pip install sleap[pypi]

Then I confirmed the installation and tried to check GPU status:

# Verify SLEAP installation
!/sleapvenv/bin/pip show sleap

# Check GPU status
!/sleapvenv/bin/python -c "import sleap; sleap.system_summary()"

The final output shows that no GPU is detected and TensorFlow throws a series of errors about missing CUDA libraries:

W tensorflow/stream_executor/platform/default/dso_loader:64] 
Could not load dynamic library 'libcudart.so.11.0' ......
Skipping registering GPU devices...
GPUs: None detected.

How can I fix this and get SLEAP to recognize and use the Colab GPU?