In [1]: import torch

In [2]: a = torch.empty((256, 512), requires_grad=True).unsqueeze(0)

In [3]: b = torch.empty((4, 128, 512), requires_grad=True).transpose(-1, -2)

In [4]: c = torch.empty((256, 4, 128)).movedim(1, 0)

In [5]: _ = torch.matmul(a.detach(), b.detach(), out=c)

In [6]: _ = torch.matmul(a, b, out=c)
---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
Cell In[11], line 1
----> 1 _ = torch.matmul(a, b, out=c)

RuntimeError: !(transpose && t2_is_matrix) INTERNAL ASSERT FAILED at "/opt/conda/conda-bld/pytorch_1695392035629/work/aten/src/ATen/native/LinearAlgebra.cpp":2026, please report a bug to PyTorch.