diff --git a/.circleci/config.yml b/.circleci/config.yml
index 78358d11..65e392d2 100644
--- a/.circleci/config.yml
+++ b/.circleci/config.yml
@@ -24,7 +24,7 @@ jobs:
             - checkout
             - run: sudo pip install --progress-bar off .
             - run: sudo pip install pytest codecov pytest-cov
-            - run: sudo pip install tensorboardX scikit-learn
+            - run: sudo pip install tensorboardX scikit-learn mock
             - run: python -m pytest -sv ./pytorch_transformers/tests/ --cov
             - run: python -m pytest -sv ./examples/
             - run: codecov
diff --git a/pytorch_transformers/modeling_gpt2.py b/pytorch_transformers/modeling_gpt2.py
index 66ff4e71..29d1cbae 100644
--- a/pytorch_transformers/modeling_gpt2.py
+++ b/pytorch_transformers/modeling_gpt2.py
@@ -423,7 +423,7 @@ class GPT2PreTrainedModel(PreTrainedModel):
         """
         num_special_tokens = kwargs.pop('num_special_tokens', None)
 
-        model = PreTrainedModel.from_pretrained(cls, pretrained_model_name_or_path, *inputs, **kwargs)
+        model = super(PreTrainedModel, cls).from_pretrained(pretrained_model_name_or_path, *inputs, **kwargs)
 
         # Add additional embeddings for special tokens if needed
         # This step also make sure we are still sharing the output and input embeddings after loading weights
diff --git a/pytorch_transformers/modeling_openai.py b/pytorch_transformers/modeling_openai.py
index c81d082c..aa35b163 100644
--- a/pytorch_transformers/modeling_openai.py
+++ b/pytorch_transformers/modeling_openai.py
@@ -431,7 +431,7 @@ class OpenAIGPTPreTrainedModel(PreTrainedModel):
         num_special_tokens = kwargs.get('num_special_tokens', None)
         kwargs.pop('num_special_tokens', None)
 
-        model = PreTrainedModel.from_pretrained(cls, pretrained_model_name_or_path, *inputs, **kwargs)
+        model = super(PreTrainedModel, cls).from_pretrained(pretrained_model_name_or_path, pretrained_model_name_or_path, *inputs, **kwargs)
 
         # Add additional embeddings for special tokens if needed
         # This step also make sure we are still sharing the output and input embeddings after loading weights