解决读取bert权重潜在的bug问题

fastNLP/embeddings/bert_embedding.py

@@ -63,7 +63,7 @@ class BertEmbedding(ContextualEmbedding):
             model_dir = cached_path(model_url)
             # 检查是否存在
         elif os.path.isdir(os.path.expanduser(os.path.abspath(model_dir_or_name))):
-            model_dir = model_dir_or_name
+            model_dir = os.path.expanduser(os.path.abspath(model_dir_or_name))
         else:
             raise ValueError(f"Cannot recognize {model_dir_or_name}.")
 

fastNLP/modules/utils.py

@@ -128,9 +128,9 @@ def _get_file_name_base_on_postfix(dir_path, postfix):
     :param postfix: 形如".bin", ".json"等
     :return: str，文件的路径
     """
-    files = glob.glob(os.path.join(dir_path, '*' + postfix))
+    files = list(filter(lambda filename:filename.endswith(postfix), os.listdir(os.path.join(dir_path))))
     if len(files) == 0:
-        raise FileNotFoundError(f"There is no file endswith *.{postfix} file in {dir_path}")
+        raise FileNotFoundError(f"There is no file endswith *{postfix} file in {dir_path}")
     elif len(files) > 1:
-        raise FileExistsError(f"There are multiple *.{postfix} files in {dir_path}")
+        raise FileExistsError(f"There are multiple *{postfix} files in {dir_path}")
     return os.path.join(dir_path, files[0])