BUG-Fixed: [CT][MS][Document] correct the description of the operator sequence_mask

mindspore/ops/composite/array_ops.py

@@ -113,11 +113,12 @@ def sequence_mask(lengths, maxlen):
     [d_1, d_2, ..., d_n, maxlen], with mask[i_1, i_2, ..., i_n, j] = (j < lengths[i_1, i_2, ..., i_n])
 
     Inputs:
-        - **lengths** (Tensor) - Tensor to calculate the mask for. All values in this tensor must be
-          less than or equal to `maxlen`. Must be type int32 or int64.
+        - **lengths** (Tensor) - Tensor to calculate the mask for. All values in this tensor should be
+          less than or equal to `maxlen`. Values greater than `maxlen` will be treated as `maxlen`.
+          Must be type int32 or int64.
 
         - **maxlen** (int) - size of the last dimension of returned tensor. Must be positive and same
-          type as elements in `lengths`. Default is the maximum value in lengths.
+          type as elements in `lengths`.
 
     Outputs:
         One mask tensor of shape lengths.shape + (maxlen,).

mindspore/ops/operations/_inner_ops.py

@@ -689,11 +689,12 @@ class SequenceMask(PrimitiveWithCheck):
     [d_1, d_2, ..., d_n, maxlen], with mask[i_1, i_2, ..., i_n, j] = (j < lengths[i_1, i_2, ..., i_n])
 
     Inputs:
-        - **lengths** (Tensor) - Tensor to calculate the mask for. All values in this tensor must be
-          less than or equal to `maxlen`. Must be type int32 or int64.
+        - **lengths** (Tensor) - Tensor to calculate the mask for. All values in this tensor should be
+          less than or equal to `maxlen`. Values greater than `maxlen` will be treated as `maxlen`.
+          Must be type int32 or int64.
 
         - **maxlen** (int) - size of the last dimension of returned tensor. Must be positive and same
-          type as elements in `lengths`. Default is the maximum value in lengths.
+          type as elements in `lengths`.
 
     Outputs:
         One mask tensor of shape lengths.shape + (maxlen,).