Traceback (most recent call last): File "bug.py", line 7, in <module> view(torch.empty(1, 1)) ~~~~^^^^^^^^^^^^^^^^^^^ File "/.../lib/python3.14/site-packages/torch/_functorch/apis.py", line 220, in wrapped return vmap_impl( # pyrefly: ignore[bad-argument-type] ...<6 lines>... **kwargs, ) File "/.../lib/python3.14/site-packages/torch/_functorch/vmap.py", line 316, in vmap_impl return _flat_vmap( func, ...<6 lines>... **kwargs, ) File "/.../lib/python3.14/site-packages/torch/_functorch/vmap.py", line 507, in _flat_vmap batched_outputs = func(*batched_inputs, **kwargs) File "bug.py", line 5, in view return t.view(torch.uint8) ~~~~~~^^^^^^^^^^^^^ RuntimeError: Batching rule not implemented for aten::view.dtype; the fallback path doesn't work on out= or view ops.

TL;DR

The issue is likely due to the torch.vmap decorator not supporting the view method with a dtype argument, and a workaround is to remove the torch.vmap decorator or modify the view method to not use the dtype argument.

Guidance

• The error message indicates that the batching rule is not implemented for the aten::view.dtype operation, which suggests that the issue is related to the torch.vmap decorator.
• Try removing the torch.vmap decorator from the view function to see if it resolves the issue.
• If the torch.vmap decorator is necessary, consider modifying the view method to not use the dtype argument, or use a different method that is supported by torch.vmap.
• Verify that the issue is resolved by running the code without the torch.vmap decorator or with the modified view method.

Example

import torch

def view(t: torch.Tensor) -> torch.Tensor:
    return t.view(torch.uint8)

view(torch.empty(1, 1))

Notes

The issue may be specific to the version of PyTorch being used (2.13.0.dev20260417), and it's possible that the issue has been resolved in a later version.

Recommendation

Apply workaround: Remove the torch.vmap decorator or modify the view method to not use the dtype argument, as this is the most straightforward way to resolve the issue without upgrading to a potentially unstable version of PyTorch.