Update on "Remove warnings on non-buffer tensor constants"

tugsbayasgalan · tugsbayasgalan · commit 9a3b4447289e · 2025-03-11T13:38:05.000-07:00
Export already registers tensor constants directly in the graph and this is also true for Torchbind objects. This removes warning that pollutes the output. cc ezyang SherlockNoMad EikanWang jgong5 wenzhe-nrv Differential Revision: [D70577856](https://our.internmc.facebook.com/intern/diff/D70577856) [ghstack-poisoned]
diff --git a/torch/fx/graph.py b/torch/fx/graph.py
@@ -1779,7 +1779,7 @@ def check_arg(arg: Node, n: Optional[Node] = None) -> None:
                                 f"Node {node} target {node.target} {atom} of {seen_qualname} does "
                                 "not reference an nn.Module"
                             )
-    
+
                         m_itr = new_m_itr
 
     @compatibility(is_backward_compatible=True)

Original file line number	Diff line number	Diff line change
`@@ -1779,7 +1779,7 @@ def check_arg(arg: Node, n: Optional[Node] = None) -> None:`
`1779`	`1779`	`f"Node {node} target {node.target} {atom} of {seen_qualname} does "`
`1780`	`1780`	`"not reference an nn.Module"`
`1781`	`1781`	`)`
`1782`		`-`
	`1782`	`+`
`1783`	`1783`	`m_itr = new_m_itr`
`1784`	`1784`
`1785`	`1785`	`@compatibility(is_backward_compatible=True)`