: python/monarch: pytorch InputBuffer::add() signature changed (#1739)

shayne-fletcher · facebook-github-bot · commit 5d8c0f734b22 · 2025-11-01T10:35:56.000-07:00
Summary:


pytorch `InputBuffer::add` API changed. adapt `monarch/python/monarch/_gradient_generator.cpp`
```
  realInputBuffer(node).add(
      input_nr,
      check_and_reduce(node-&gt;node, input_nr, std::move(t)),
      std::nullopt,
      std::nullopt);  // Remove: node-&gt;node
```

that gradient computation still works correctly after this change needs to be tested.

Differential Revision: D86025607
diff --git a/python/monarch/gradient/_gradient_generator.cpp b/python/monarch/gradient/_gradient_generator.cpp
@@ -424,8 +424,7 @@ struct GradientGenerator {
         input_nr,
         check_and_reduce(node->node, input_nr, std::move(t)),
         std::nullopt,
-        std::nullopt,
-        node->node);
+        std::nullopt);
   }
 
   InputBuffer& realInputBuffer(NodeState* state) {

Original file line number	Diff line number	Diff line change
`@@ -424,8 +424,7 @@ struct GradientGenerator {`
`424`	`424`	`input_nr,`
`425`	`425`	`check_and_reduce(node->node, input_nr, std::move(t)),`
`426`	`426`	`std::nullopt,`
`427`		`- std::nullopt,`
`428`		`- node->node);`
	`427`	`+ std::nullopt);`
`429`	`428`	`}`
`430`	`429`
`431`	`430`	`InputBuffer& realInputBuffer(NodeState* state) {`