Short name describing what triggered the graph break
Unsupported autograd.Function context save_for_backward
Values or code snippet captured at the break point
call_method {self} {name}
Explanation of why the graph break was triggered
Dynamo requires the saved_tensors
attribute to be initialized on the autograd.Function
context object.
Hints on how to resolve the graph break
saved_tensors
attribute is properlysave_for_backward
.save_for_backward
only supported on a newly constructed torch.autograd.function.FunctionCtx
.