add docs about annotating variables

Sidney Zhang · Sidney Zhang · commit e7ec5ba6ee4c · 2019-01-24T10:18:34.000-05:00
diff --git a/docs/stable/_sources/jit.rst.txt b/docs/stable/_sources/jit.rst.txt
@@ -184,8 +184,13 @@ Example::
                  # and type int in the false branch
 
 By default, all parameters to a Torch Script function are assumed to be Tensor
-because this is the most common type used in modules. To specify that an
-argument to a Torch Script function is another type, it is possible to use
+because this is the most common type used in modules.
+
+There are 2 scenarios in which you might want to annotate a type:
+
+1. Annotating Function Argument Types
+
+To specify that an argument to a Torch Script function is another type, it is possible to use
 MyPy-style type annotations using the types listed above:
 
 Example::
@@ -203,6 +208,32 @@ Example::
   In our examples, we use comment-based annotations to ensure Python 2
   compatibility as well.
 
+2. Annotating Variable Types
+
+For example, a list by default is assumed to be List[Tensor]. If you would like to
+have a list of other types. PyTorch provides annotation functions.
+
+Example::
+    import torch
+    from torch.jit import Tensor
+    from typing import List, Tuple
+
+    class ListOfTupleOfTensor(torch.jit.ScriptModule):
+        def __init__(self):
+            super(ListOfTupleOfTensor, self).__init__()
+
+        @torch.jit.script_method
+        def forward(self, x):
+            # type: (Tensor) -> List[Tuple[Tensor, Tensor]]
+
+            # This annotates the list to be a List[Tuple[Tensor, Tensor]]
+            returns = torch.jit.annotate(List[Tuple[Tensor, Tensor]], [])
+            for i in range(10):
+                returns.append((x, x))
+
+            return returns
+
+
 Expressions
 ~~~~~~~~~~~
 
