[Wav2Vec2] Fix convert (huggingface#11562)

* push

* small change

* correct other typo

src/transformers/models/wav2vec2/convert_wav2vec2_original_pytorch_checkpoint_to_pytorch.py

@@ -178,9 +178,11 @@ def convert_wav2vec2_checkpoint(
         if dict_path:
             target_dict = Dictionary.load(dict_path)
 
-            config.bos_token_id = target_dict.bos_index
+            # important change bos & pad token id since CTC symbol is <pad> and
+            # not <s> as in fairseq
+            config.bos_token_id = target_dict.pad_index
+            config.pad_token_id = target_dict.bos_index
             config.eos_token_id = target_dict.eos_index
-            config.pad_token_id = target_dict.pad_index
             config.vocab_size = len(target_dict.symbols)
             vocab_path = os.path.join(pytorch_dump_folder_path, "vocab.json")
             if not os.path.isdir(pytorch_dump_folder_path):
@@ -214,9 +216,8 @@ def convert_wav2vec2_checkpoint(
         hf_wav2vec = Wav2Vec2Model(config)
 
     if is_finetuned:
-
         model, _, _ = fairseq.checkpoint_utils.load_model_ensemble_and_task(
-            [checkpoint_path], arg_overrides={"data": dict_path}
+            [checkpoint_path], arg_overrides={"data": "/".join(dict_path.split("/")[:-1])}
         )
     else:
         model, _, _ = fairseq.checkpoint_utils.load_model_ensemble_and_task([checkpoint_path])