diff --git a/backends/cadence/aot/ops_registrations.py b/backends/cadence/aot/ops_registrations.py
index e73de6ab7ce..c69aaa8091f 100644
--- a/backends/cadence/aot/ops_registrations.py
+++ b/backends/cadence/aot/ops_registrations.py
@@ -164,7 +164,7 @@ def quantized_layer_norm_meta(
     output_scale: float,
     output_zero_point: int,
 ) -> torch.Tensor:
-    return input.new_empty(input.size(), dtype=torch.uint8)
+    return input.new_empty(input.size(), dtype=input.dtype)
 
 
 @register_fake("cadence::quantized_relu")
diff --git a/backends/cadence/reference/operators/quantized_matmul_out.cpp b/backends/cadence/reference/operators/quantized_matmul_out.cpp
index b0a9393cd01..5e1a49c0b5a 100644
--- a/backends/cadence/reference/operators/quantized_matmul_out.cpp
+++ b/backends/cadence/reference/operators/quantized_matmul_out.cpp
@@ -49,7 +49,7 @@ __attribute__((noinline)) void qmatmul(
           sum += (X[i * n + k] - X_zero_point) * (y[k * p + j] - Y_zero_point);
         }
       }
-      Z[i * p + j] = kernels::quantize<uint8_t>(sum, Z_scale, Z_zero_point);
+      Z[i * p + j] = kernels::quantize<TZ>(sum, Z_scale, Z_zero_point);
     }
   }
 }