You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Describe the bug
When using VIPosterior with "mps" as GPU device, and using q="nsf", and num_dim=1, then sampling from q will result in NaNs.
To Reproduce
Please add a minimal code example that reproduces the problem:
Python 3.11, SBI 0.22.0
On MacOS with MPS enabled, run pytest tests/inference_on_device_test.py::test_vi_on_gpu
Expected behavior
Sampling q should not result in NaNs.
Additional context
This issue occurs only on the MPS device and not on the CPU.
Might be related to / blocked by this PyTorch issue: pytorch/pytorch#89127
Describe the bug
When using VIPosterior with "mps" as GPU device, and using
q="nsf"
, andnum_dim=1
, then sampling fromq
will result in NaNs.To Reproduce
Please add a minimal code example that reproduces the problem:
pytest tests/inference_on_device_test.py::test_vi_on_gpu
Expected behavior
Sampling
q
should not result in NaNs.Additional context
This issue occurs only on the MPS device and not on the CPU.
Might be related to / blocked by this PyTorch issue: pytorch/pytorch#89127
see also #912
The text was updated successfully, but these errors were encountered: