[Doc clarification] Added unidirectional text for LayerNorm (#5686)

### Description Doc only, added clarification that scale and B can be uni-directionally expanded to X. Current def is not explicit on this point. ### Motivation and Context Clarification of docs to be more explicit. Issue discussed here #5666 --------- Signed-off-by: Alexandre Eichenberger <alexe@us.ibm.com> Signed-off-by: Justin Chu <justinchuby@users.noreply.github.com> Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Justin Chu <justinchuby@users.noreply.github.com>
onnx · Oct 19, 2023 · bbe7011 · bbe7011
1 parent 42b0662
commit bbe7011
Show file tree

Hide file tree

Showing 3 changed files with 9 additions and 3 deletions.
diff --git a/docs/Changelog.md b/docs/Changelog.md
@@ -21043,7 +21043,9 @@ This version of the operator has been available since version 17 of the default
         Let `d[i]` indicate the i-th dimension of `X`.
         If `X`'s shape is `[d[0], ..., d[axis-1], d[axis], ..., d[rank-1]]`,
         the shape of `Mean` and `InvStdDev` is `[d[0], ..., d[axis-1], 1, ..., 1]`.
-        `Y` and `X` have the same shape.
+        `Y` and `X` have the same shape. This operator supports unidirectional broadcasting
+        (tensors `Scale` and `B` should be unidirectional broadcastable to tensor `X`);
+        for more details please check [the doc](Broadcasting.md).
 
 #### Version
 

diff --git a/docs/Operators.md b/docs/Operators.md
@@ -13172,7 +13172,9 @@ expect(
         Let `d[i]` indicate the i-th dimension of `X`.
         If `X`'s shape is `[d[0], ..., d[axis-1], d[axis], ..., d[rank-1]]`,
         the shape of `Mean` and `InvStdDev` is `[d[0], ..., d[axis-1], 1, ..., 1]`.
-        `Y` and `X` have the same shape.
+        `Y` and `X` have the same shape. This operator supports unidirectional broadcasting
+        (tensors `Scale` and `B` should be unidirectional broadcastable to tensor `X`);
+        for more details please check [the doc](Broadcasting.md).
 
 #### Version
 

diff --git a/onnx/defs/nn/defs.cc b/onnx/defs/nn/defs.cc
@@ -2516,7 +2516,9 @@ static const char* LayerNormalization_ver17_doc = R"DOC(
       Let `d[i]` indicate the i-th dimension of `X`.
       If `X`'s shape is `[d[0], ..., d[axis-1], d[axis], ..., d[rank-1]]`,
       the shape of `Mean` and `InvStdDev` is `[d[0], ..., d[axis-1], 1, ..., 1]`.
-      `Y` and `X` have the same shape.
+      `Y` and `X` have the same shape. This operator supports unidirectional broadcasting
+      (tensors `Scale` and `B` should be unidirectional broadcastable to tensor `X`);
+      for more details please check [the doc](Broadcasting.md).
 )DOC";
 
 bool BuildContextDependentFunctionBodyLayerNormalization(