diff --git a/demo/finetune_classifier_distributed.py b/demo/finetune_classifier_distributed.py
index 3dfa2f680c1a82a032495911d30bbbb7a52e562b..d1df8675369125baf3c1605347cfa0848b0903de 100644
--- a/demo/finetune_classifier_distributed.py
+++ b/demo/finetune_classifier_distributed.py
@@ -128,7 +128,7 @@ model = ErnieModelForSequenceClassification.from_pretrained(
 
 if args.init_checkpoint is not None:
     log.info('loading checkpoint from %s' % args.init_checkpoint)
-    sd, _ = P.load(args.init_checkpoint)
+    sd = P.load(args.init_checkpoint)
     model.set_state_dict(sd)
 
 model = P.DataParallel(model)
diff --git a/demo/finetune_sentiment_analysis.py b/demo/finetune_sentiment_analysis.py
index 794619840926305ba8de55342a77e00cb2cb821d..015d29d4d2fe0acdf34d0032ebdffd3946515e62 100644
--- a/demo/finetune_sentiment_analysis.py
+++ b/demo/finetune_sentiment_analysis.py
@@ -34,7 +34,6 @@ import propeller.paddle as propeller
 
 log.setLevel(logging.DEBUG)
 logging.getLogger().setLevel(logging.DEBUG)
-log = logging.getLogger()
 
 #from model.bert import BertConfig, BertModelLayer
 from ernie.modeling_ernie import ErnieModel, ErnieModelForSequenceClassification
@@ -70,6 +69,11 @@ parser.add_argument('--lr', type=float, default=5e-5, help='learning rate')
 parser.add_argument('--eval', action='store_true')
 parser.add_argument(
     '--save_dir', type=Path, required=True, help='model output directory')
+parser.add_argument(
+    '--init_checkpoint',
+    type=str,
+    default=None,
+    help='checkpoint to warm start from')
 parser.add_argument(
     '--wd', type=float, default=0.01, help='weight decay, aka L2 regularizer')
 parser.add_argument(
@@ -185,7 +189,7 @@ else:
             tokenizer=tokenizer.tokenize),
     ])
 
-    sd, _ = P.load(args.save_dir / 'ckpt.bin')
+    sd = P.load(args.init_checkpoint)
     model.set_dict(sd)
     model.eval()
 
diff --git a/ernie/modeling_ernie.py b/ernie/modeling_ernie.py
index b8b5e8ebce75c6e49e97cccf05fadd08bae2d90d..2d0637c88fb147868fa7dc460d540bf4825c8389 100644
--- a/ernie/modeling_ernie.py
+++ b/ernie/modeling_ernie.py
@@ -230,12 +230,14 @@ class PretrainedModel(object):
                         pretrain_dir_or_url,
                         force_download=False,
                         **kwargs):
-        if not Path(pretrain_dir_or_url).exists() and pretrain_dir_or_url in cls.resource_map:
-            url = cls.resource_map[pretrain_dir_or_url]
+        if not Path(pretrain_dir_or_url).exists() and str(
+                pretrain_dir_or_url) in cls.resource_map:
+            url = cls.resource_map[str(pretrain_dir_or_url)]
             log.info('get pretrain dir from %s' % url)
             pretrain_dir = _fetch_from_remote(url, force_download)
         else:
-            log.info('pretrain dir %s not in %s, read from local' % (pretrain_dir_or_url, repr(cls.resource_map)))
+            log.info('pretrain dir %s not in %s, read from local' %
+                     (pretrain_dir_or_url, repr(cls.resource_map)))
             pretrain_dir = Path(pretrain_dir_or_url)
 
         if not pretrain_dir.exists():
diff --git a/ernie/tokenizing_ernie.py b/ernie/tokenizing_ernie.py
index ffcefd2eb943d451c68069e372f0148ceb188515..b4cc247eeba113fc8d62bd63197c3d398922ac90 100644
--- a/ernie/tokenizing_ernie.py
+++ b/ernie/tokenizing_ernie.py
@@ -94,16 +94,16 @@ class ErnieTokenizer(object):
                         pretrain_dir_or_url,
                         force_download=False,
                         **kwargs):
-        if pretrain_dir_or_url in cls.resource_map:
-            url = cls.resource_map[pretrain_dir_or_url]
+        if not Path(pretrain_dir_or_url).exists() and str(
+                pretrain_dir_or_url) in cls.resource_map:
+            url = cls.resource_map[str(pretrain_dir_or_url)]
             log.info('get pretrain dir from %s' % url)
             pretrain_dir = _fetch_from_remote(
                 url, force_download=force_download)
         else:
             log.info('pretrain dir %s not in %s, read from local' %
                      (pretrain_dir_or_url, repr(cls.resource_map)))
-            pretrain_dir = pretrain_dir_or_url
-        pretrain_dir = Path(pretrain_dir)
+            pretrain_dir = Path(pretrain_dir_or_url)
         if not pretrain_dir.exists():
             raise ValueError('pretrain dir not found: %s' % pretrain_dir)
         vocab_path = pretrain_dir / 'vocab.txt'
@@ -235,12 +235,14 @@ class ErnieTinyTokenizer(ErnieTokenizer):
                         pretrain_dir_or_url,
                         force_download=False,
                         **kwargs):
-        if pretrain_dir_or_url in cls.resource_map:
-            url = cls.resource_map[pretrain_dir_or_url]
+        if not Path(pretrain_dir_or_url).exists() and str(
+                pretrain_dir_or_url) in cls.resource_map:
+            url = cls.resource_map[str(pretrain_dir_or_url)]
             log.info('get pretrain dir from %s' % url)
             pretrain_dir = _fetch_from_remote(url, force_download)
         else:
-            log.info('pretrain dir %s not in %s, read from local' % (pretrain_dir_or_url, repr(cls.resource_map)))
+            log.info('pretrain dir %s not in %s, read from local' %
+                     (pretrain_dir_or_url, repr(cls.resource_map)))
             pretrain_dir = Path(pretrain_dir_or_url)
         if not pretrain_dir.exists():
             raise ValueError('pretrain dir not found: %s' % pretrain_dir)