Merge pull request #2005 from guoshengCS/fix-transformer-core

Remove core api in Transformer

PaddleNLP/neural_machine_translation/transformer/infer.py

@@ -281,10 +281,10 @@ def fast_infer(args):
                     if feed_dict_list is not None else None,
                     return_numpy=False,
                     use_program_cache=True)
-            seq_ids_list, seq_scores_list = [seq_ids], [
-                seq_scores
-            ] if isinstance(
-                seq_ids, paddle.fluid.core.LoDTensor) else (seq_ids, seq_scores)
+            seq_ids_list, seq_scores_list = [
+                seq_ids
+            ], [seq_scores] if isinstance(
+                seq_ids, paddle.fluid.LoDTensor) else (seq_ids, seq_scores)
             for seq_ids, seq_scores in zip(seq_ids_list, seq_scores_list):
                 # How to parse the results:
                 #   Suppose the lod of seq_ids is:

PaddleNLP/neural_machine_translation/transformer/model.py

@@ -875,7 +875,7 @@ def fast_decode(src_vocab_size,
             accu_scores = layers.elementwise_add(
                 x=layers.log(topk_scores), y=pre_scores, axis=0)
             # beam_search op uses lod to differentiate branches.
-            topk_indices = layers.lod_reset(topk_indices, pre_ids)
+            topk_indices = layers.lod_reset(accu_scores, pre_ids)
             # topK reduction across beams, also contain special handle of
             # end beams and end sentences(batch reduction)
             selected_ids, selected_scores, gather_idx = layers.beam_search(