Dynamo doesn't support to(..., non_blocking=True) (pytorch#1979)

Summary:
Pull Request resolved: pytorch#1979

as per title.

Also internal workstreams are registering is_torchdynamo_compiling() as True in some workstreams, seems like a seperate bug.

Reviewed By: IvanKobzarev

Differential Revision:
D57190750

Privacy Context Container: 1203980333745195

fbshipit-source-id: 1ecb45040658ca3d4998d4efe1fe2c170e88065d

torchrec/sparse/jagged_tensor.py

@@ -63,8 +63,8 @@ def is_non_strict_exporting() -> bool:
 
 def _pin_and_move(tensor: torch.Tensor, device: torch.device) -> torch.Tensor:
     if is_torchdynamo_compiling():
-        # TODO(ivankobzarev): Dynamo trace with pin_memory once FakeTensor supports it
-        return tensor.to(device=device, non_blocking=True)
+        # TODO: remove once FakeTensor supports pin_memory() and to(..., non_blocking=True)
+        return tensor.to(device=device)
 
     return (
         tensor.pin_memory().to(device=device, non_blocking=True)