add layerNorm GB national standard operator's explanation

diopi_test/python/configs/diopi_configs.py

@@ -7986,6 +7986,7 @@
         ),
     ),
 
+    #This test config is for the national standard GB operator version of diopiLayerNorm, which is different from the original interface definition.
     'layer_norm': dict(
         name=["layer_norm"],
         dtype=[np.float32, np.float64, np.float16],

diopi_test/python/conformance/diopi_functions.py

@@ -5270,6 +5270,8 @@ def layer_normGB(input, normalized_shape, weight=None, bias=None, eps=1e-05):
     bias = None if bias is None else bias
 
     out = raw_like(input)
+
+    # Note that, this is the national standard GB operator version, which is different from the diopiLayerNorm interface definition, normalized_shape has changed to begin_norm_axis.
     func = check_function("diopiLayerNormGB")
     ret = func(
         input.context(),

proto/include/diopi/functions.h

@@ -3488,7 +3488,8 @@ DIOPI_API diopiError_t diopiNormBackward(diopiContextHandle_t ctx, diopiTensorHa
 
 /**
  * @brief Applies Layer Normalization over a mini-batch of inputs.
- * type=[float32, float64, float16].
+ * Note that, this is the national standard GB operator version, which is different from the diopiLayerNorm interface definition, normalized_shape has changed
+ * to begin_norm_axis. type=[float32, float64, float16].
  * @param[in] ctx Context environment.
  * @param[in] save_mean Mean tensor,the mean value for each feature channel of the input tensor. type=[float32, float64, float16].
  * @param[in] save_invstd Backup of inverse standard deviation computed during training. type=[float32, float64, float16].
@@ -3933,6 +3934,8 @@ DIOPI_API diopiError_t diopiLayerNormBackward(diopiContextHandle_t ctx, diopiTen
 
 /**
  * @brief Compute the backward pass for diopiLayerNormGB(). Computes gradients for input, weight, and bias.
+ * Note that, this is the national standard GB operator version, which is different from the diopiLayerNormBackward interface definition, normalized_shape has
+ * changed to begin_norm_axis
  * @param[in] ctx Context environment.
  * @param[in] grad_output the grad tensor of output. type=[float32, float64, float16].
  * @param[in] grad_bias the grad of bias. type=[float32, float64, float16].