Add tracks_own_finished to Decoder to avoid mismanagement of the finished state in dynamic_decode. (#23664)

test=develop

python/paddle/fluid/layers/rnn.py

@@ -601,6 +601,28 @@ class Decoder(object):
         """
         raise NotImplementedError
 
+    @property
+    def tracks_own_finished(self):
+        """
+        Describes whether the Decoder keeps track of finished states by itself.
+
+        `decoder.step()` would emit a bool `finished` value at each decoding
+        step. The emited `finished` can be used to determine whether every
+        batch entries is finished directly, or it can be combined with the
+        finished tracker keeped in `dynamic_decode` by performing a logical OR
+        to take the already finished into account.
+
+        If `False`, the latter would be took when performing `dynamic_decode`,
+        which is the default. Otherwise, the former would be took, which uses
+        the finished value emited by the decoder as all batch entry finished
+        status directly, and it is the case when batch entries might be
+        reordered such as beams in BeamSearchDecoder.
+
+        Returns:
+            bool: A python bool `False`.
+        """
+        return False
+
 
 class BeamSearchDecoder(Decoder):
     """
@@ -1048,6 +1070,19 @@ class BeamSearchDecoder(Decoder):
         # TODO: use FinalBeamSearchDecoderOutput as output
         return predicted_ids, final_states
 
+    @property
+    def tracks_own_finished(self):
+        """
+        BeamSearchDecoder reorders its beams and their finished state. Thus it
+        conflicts with `dynamic_decode` function's tracking of finished states.
+        Setting this property to true to avoid early stopping of decoding due
+        to mismanagement of the finished state.
+
+        Returns:
+            bool: A python bool `True`.
+        """
+        return True
+
 
 def dynamic_decode(decoder,
                    inits=None,
@@ -1205,7 +1240,13 @@ def dynamic_decode(decoder,
                 states_arrays)
         (outputs, next_states, next_inputs,
          next_finished) = decoder.step(step_idx, inputs, states, **kwargs)
-        next_finished = control_flow.logical_or(next_finished, global_finished)
+        if not decoder.tracks_own_finished:
+            # BeamSearchDecoder would track it own finished, since beams would
+            # be reordered and the finished status of each entry might change.
+            # Otherwise, perform logical OR which would not change the already
+            # finished.
+            next_finished = control_flow.logical_or(next_finished,
+                                                    global_finished)
         next_sequence_lengths = nn.elementwise_add(
             sequence_lengths,
             tensor.cast(
@@ -1226,6 +1267,10 @@ def dynamic_decode(decoder,
             lambda x, x_array: control_flow.array_write(
                 x, i=step_idx, array=x_array), outputs, outputs_arrays)
         control_flow.increment(x=step_idx, value=1.0, in_place=True)
+        # update the global_finished first, since it might be also in states of
+        # decoder, which otherwise would write a stale finished status to array
+        tensor.assign(next_finished, global_finished)
+        tensor.assign(next_sequence_lengths, sequence_lengths)
         if is_test:
             map_structure(tensor.assign, next_inputs, global_inputs)
             map_structure(tensor.assign, next_states, global_states)
@@ -1236,8 +1281,6 @@ def dynamic_decode(decoder,
             map_structure(
                 lambda x, x_array: control_flow.array_write(
                     x, i=step_idx, array=x_array), next_states, states_arrays)
-        tensor.assign(next_finished, global_finished)
-        tensor.assign(next_sequence_lengths, sequence_lengths)
         if max_step_num is not None:
             control_flow.logical_and(
                 control_flow.logical_not(nn.reduce_all(global_finished)),