GB7299

Graph-Break Type

Short name describing what triggered the graph break

_autograd_grad with lost grad_fn linkage

Context

Values or code snippet captured at the break point

outputs lost autograd linkage during tracing

Explanation

Explanation of why the graph break was triggered

_autograd_grad() received tensors whose grad_fn was lost during tracing - this silently produces zero gradients.

Hints

Hints on how to resolve the graph break

• Compile the full transform instead of the returned
• closure: torch.compile(lambda x: torch.func.vjp(f, x))
• It may be possible to write Dynamo tracing rules for this code. Please report an issue to PyTorch if you encounter this graph break often and it is causing performance issues.

Additional Information

Click here to add Additional Info

Back to Registry