[Dy2Stat] Fix Error when generating train_program in eval mode (#27975) (#28019)

* Fix save in eval mode

* remove assert statement

* fix test_partial_program failed

* add more test

* modify back into _train_program

python/paddle/fluid/dygraph/dygraph_to_static/partial_program.py

@@ -82,6 +82,29 @@ class NestSequence(object):
         return self.tolist()[item]
 
 
+class LazyInitialized(object):
+    """
+    Descriptor to implement lazy initialization of property.
+    """
+
+    def __init__(self, function):
+        self.function = function
+
+    def __get__(self, instance, cls):
+        val = self.function(instance)
+        setattr(instance, self.function.__name__, val)
+        return val
+
+
+def _change_is_test_status(program, is_test):
+    # change all `is_test` attributes
+    for block in program.blocks:
+        for op in block.ops:
+            if op.has_attr('is_test'):
+                op._set_attr('is_test', is_test)
+    return program
+
+
 class PartialProgramLayer(layers.Layer):
     """
     PartialProgramLayer wraps all the ops from layers decorated by `@declarative`
@@ -109,15 +132,30 @@ class PartialProgramLayer(layers.Layer):
         self._outputs = NestSequence(outputs, need_check=True)
         self._params = parameters if parameters is not None else []
 
-        main_program = self._verify_program(main_program)
-        self._infer_program = self._clone_for_test(main_program)
-        self._train_program = self._append_backward_desc(main_program)
-
-        self._set_grad_type(self._params)
+        self._origin_main_program = self._verify_program(main_program)
         self._inner_scope = core.Scope()
         # Set default mode to train
         self.training = True
 
+    @LazyInitialized
+    def _infer_program(self):
+        """
+        Lazy initialized property of infer_program.
+        """
+        return self._clone_for_test(self._origin_main_program)
+
+    @LazyInitialized
+    def _train_program(self):
+        """
+        Lazy initialized property of train_program.
+        """
+        train_program = self._append_backward_desc(self._origin_main_program)
+        # Note: Only set grad type once after initializing train program. So we
+        # put it here.
+        self._set_grad_type(self._params, train_program)
+
+        return train_program
+
     def _verify_program(self, main_program):
         """
         Verify that the program parameter is initialized, prune some unused params,
@@ -132,7 +170,8 @@ class PartialProgramLayer(layers.Layer):
 
     @switch_to_static_graph
     def _append_backward_desc(self, main_program):
-        program = main_program.clone()
+        # make sure all status of is_test are False in train mode.
+        program = _change_is_test_status(main_program.clone(), is_test=False)
         targets = []
         for out in self._outputs.tolist():
             if isinstance(out, framework.Variable):
@@ -280,7 +319,7 @@ class PartialProgramLayer(layers.Layer):
 
         return out_vars
 
-    def _set_grad_type(self, params):
+    def _set_grad_type(self, params, train_program):
         # NOTE: if user set sparse gradient mode, the param's gradient
         # will be SelectedRows, not LoDTensor. But tracer will just
         # set param grad VarBase by forward VarBase(LoDTensor)
@@ -289,7 +328,7 @@ class PartialProgramLayer(layers.Layer):
         # be user wanted result.
         for param in params:
             grad_name = param.name + core.grad_var_suffix()
-            grad_var = self._train_program.desc.block(0).find_var(
+            grad_var = train_program.desc.block(0).find_var(
                 cpt.to_bytes(grad_name))
             # NOTE: cannot find var desc maybe no problem, such as in batch_norm
             if grad_var is None:

python/paddle/fluid/tests/unittests/dygraph_to_static/test_lstm.py

@@ -51,6 +51,32 @@ class TestLstm(unittest.TestCase):
             msg='dygraph_out is {}\n static_out is \n{}'.format(dygraph_out,
                                                                 static_out))
 
+    def test_save_in_eval(self):
+        paddle.jit.ProgramTranslator().enable(True)
+        net = Net(12, 2)
+        # switch eval mode firstly
+        net.eval()
+        net = paddle.jit.to_static(
+            net, input_spec=[paddle.static.InputSpec(shape=[-1, 10, 12])])
+        paddle.jit.save(net, 'simple_lstm')
+        # load saved model
+        load_net = paddle.jit.load('simple_lstm')
+
+        x = paddle.randn((2, 10, 12))
+        dygraph_out = net(x)
+        static_out = load_net(x)
+        self.assertTrue(
+            np.allclose(dygraph_out.numpy(), static_out.numpy()),
+            msg='dygraph_out is {}\n static_out is \n{}'.format(dygraph_out,
+                                                                static_out))
+        # switch back into train mode.
+        net.train()
+        train_out = net(x)
+        self.assertTrue(
+            np.allclose(dygraph_out.numpy(), train_out.numpy()),
+            msg='dygraph_out is {}\n static_out is \n{}'.format(dygraph_out,
+                                                                train_out))
+
 
 if __name__ == "__main__":
     unittest.main()