cudaErrorNoKernelImageForDevice

[SNicolaJ]

I managed to get sleap to open, but I do not get the training runing. I got the following error:torch.AcceleratorError: CUDA error: no kernel image is available for execution on the device
Search for `cudaErrorNoKernelImageForDevice' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html for more information.
For more detailed error information, run with CUDA_LOG_FILE=stderr

How can I solve it?

SLEAP System Diagnostics

Generated: 2026-08-28 17:35:06

[Platform]
OS: Windows 11
Platform: Windows-11-10.0.26100-SP0
Machine: AMD64
Processor: Intel64 Family 6 Model 85 Stepping 7, GenuineIntel
Disk: 391.4 GB / 475.9 GB

[Python]
Version: 3.13.15
Executable: C:\Users\sjung2\AppData\Roaming\uv\tools\sleap\Scripts\python.exe
Prefix: C:\Users\sjung2\AppData\Roaming\uv\tools\sleap

[UV]
Version: uv 0.12.7 (61291a8ca 2026-08-27 x86_64-pc-windows-msvc)
Path: C:\Users\sjung2.local\bin\uv.EXE
Cache Dir: C:\Users\sjung2\AppData\Local\uv\cache
Tool Dir: C:\Users\sjung2\AppData\Roaming\uv\tools
Tool Bin Dir: C:\Users\sjung2.local\bin
Python Dir: C:\Users\sjung2\AppData\Roaming\uv\python
Installed Tools: sleap

[UV Config]
Default Python: (not configured)
Resolved Python: C:\Users\sjung2\AppData\Roaming\uv\python\cpython-3.13-windows-x86_64-none\python.exe
Python Preference: managed (default)
Resolution: highest (default)
Index Strategy: first-index (default)
Prerelease: if-necessary (default)

[GPU / CUDA]
NVIDIA Driver: 582.42 (CUDA 13.0)
GPU 0: Quadro P2200 (3846 MB free / 5120 MB)
PyTorch: v2.13.0+cu132 (CUDA 13.2)

[Packages]
sleap: v1.6.5 (pip)
sleap-io: v0.9.2 (pip)
sleap-nn: v0.3.3 (pip)
numpy: v2.5.2 (pip)
h5py: v3.16.0 (pip)
PySide6: v6.11.2 (pip)
opencv-python: v5.0.0.93 (pip)
torch: v2.13.0+cu132 (pip)
imageio: v2.37.4 (pip)
imageio-ffmpeg: v0.6.0 (pip)
av: v15.1.0 (pip)

[CLI Binaries]
sleap:
Path: C:\Users\sjung2.local\bin\sleap.EXE
Real Path: C:\Users\sjung2.local\bin\sleap.exe
Source: pip
ffmpeg (imageio):
Path:
C:\Users\sjung2\AppData\Roaming\uv\tools\sleap\Lib\site-packages\imageio_ffmpeg\binaries\ffmpeg-win-x86_64-v7.1.exe
Source: imageio-ffmpeg
Python: v7.1-essentials_build-www.gyan.dev

[PATH (relevant entries)]
C:\Users\sjung2\AppData\Roaming\uv\tools\sleap\Lib\site-packages\cv2../../x64/vc17/bin
C:\Users\sjung2.local\bin

Copy this output when reporting issues at:
https://github.com/talmolab/sleap/issues

Tip: Use sleap doctor -o to save diagnostics to a file

[tom21100227]

Hi @SNicolaJ ,

This seems like an known issue (see #2628 ) since Pascal generation GPU (the P2200 that you have) support has been dropped on PyTorch.

Downgrading to PyTorch 2.8 with CUDA 12.6 should work (uv pip install torch==2.8.0 --index-url https://download.pytorch.org/whl/cu126) but that might cause dependency issues with other sleap-nn dependencies.

Alternatively, try exporting to ONNX and run inference on CPU, although that would probably be slower since GPU inference is faster. See #2628 for instructions on that.

Hope this helps!
Tom