add the translation path to save the translation results

fairseq_cli/generate.py

@@ -43,8 +43,12 @@ def main(cfg: DictConfig):
             cfg.common_eval.results_path,
             "generate-{}.txt".format(cfg.dataset.gen_subset),
         )
+        translation_path = os.path.join(
+            cfg.common_eval.results_path,
+            "translation-{}.txt".format(cfg.dataset.gen_subset)
+        )
         with open(output_path, "w", buffering=1, encoding="utf-8") as h:
-            return _main(cfg, h)
+            return _main(cfg, h, translation_path)
     else:
         return _main(cfg, sys.stdout)
 
@@ -56,7 +60,7 @@ def get_symbols_to_strip_from_output(generator):
         return {generator.eos}
 
 
-def _main(cfg: DictConfig, output_file):
+def _main(cfg: DictConfig, output_file, translation_path=None):
     logging.basicConfig(
         format="%(asctime)s | %(levelname)s | %(name)s | %(message)s",
         datefmt="%Y-%m-%d %H:%M:%S",
@@ -179,6 +183,7 @@ def _main(cfg: DictConfig, output_file):
         return x
 
     scorer = scoring.build_scorer(cfg.scoring, tgt_dict)
+    translation_list = []
 
     num_sentences = 0
     has_target = True
@@ -233,7 +238,7 @@ def _main(cfg: DictConfig, output_file):
                     sample_id
                 )
             else:
-                if src_dict is not None and src_tokens in [torch.int32, torch.int64]:
+                if src_dict is not None and src_tokens.dtype in [torch.int32, torch.int64]:
                     src_str = src_dict.string(src_tokens, cfg.common_eval.post_process)
                 else:
                     src_str = ""
@@ -347,6 +352,7 @@ def _main(cfg: DictConfig, output_file):
 
                 # Score only the top hypothesis
                 if has_target and j == 0:
+                    translation_list.append([src_str, target_str, detok_hypo_str])
                     if align_dict is not None or cfg.common_eval.post_process is not None:
                         # Convert back to tokens for evaluation with unk replacement and/or without BPE
                         target_tokens = tgt_dict.encode_line(
@@ -393,6 +399,11 @@ def _main(cfg: DictConfig, output_file):
             ),
             file=output_file,
         )
+        if translation_path is not None:
+            with open(translation_path, "w", encoding="utf-8") as f:
+                for item in translation_list:
+                    f.write("{}\n".format("\t".join(item)))
+
 
     return scorer