From a4d848277ce5695a61826c4ea44c447b24082bd7 Mon Sep 17 00:00:00 2001
From: Satya Kumar Jandhyala <satya.k.jandhyala@gmail.com>
Date: Tue, 21 Jan 2025 14:32:49 -0800
Subject: [PATCH] Fix the input_output_stride for 4-dim input.

---
 onnxruntime/contrib_ops/webgpu/bert/rotary_embedding.cc | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/onnxruntime/contrib_ops/webgpu/bert/rotary_embedding.cc b/onnxruntime/contrib_ops/webgpu/bert/rotary_embedding.cc
index bc8b7493fc..5e0703f87f 100644
--- a/onnxruntime/contrib_ops/webgpu/bert/rotary_embedding.cc
+++ b/onnxruntime/contrib_ops/webgpu/bert/rotary_embedding.cc
@@ -95,7 +95,7 @@ Status RotaryEmbedding::ComputeInternal(onnxruntime::webgpu::ComputeContext& con
       input_shape.NumDimensions() == 3
           ? std::vector<uint32_t>({batch_stride, hidden_size, head_size, 1})
           : (input_shape.NumDimensions() == 4
-                 ? std::vector<uint32_t>({batch_stride, head_size, sequence_length * head_size, 1})
+                 ? std::vector<uint32_t>({batch_stride, sequence_length * head_size, head_size, 1})
                  : std::vector<uint32_t>({}));
 
   program