GB8985

Graph-Break Type

Short name describing what triggered the graph break

requires_grad_() intermediate leaked as output

Context

Values or code snippet captured at the break point

call_method {self} requires_grad_

Explanation

Explanation of why the graph break was triggered

An intermediate tensor with requires_grad_() called on it (or a tensor derived from it) is returned from the compiled region. Graph breaking here to preserve partial acceleration.

Hints

Hints on how to resolve the graph break

• Call .detach() before returning if you only need values.
• Consume the gradient inside the compiled function
• (call backward() and use .grad),
• or move requires_grad_() outside torch.compile.

Additional Information

Click here to add Additional Info

Back to Registry