Encode and Decode are back in the superclass. They now handle sentence pairs special tokens.

pytorch_transformers/modeling_roberta.py

@@ -23,7 +23,7 @@ import logging
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
-from torch.nn import CrossEntropyLoss
+from torch.nn import CrossEntropyLoss, MSELoss
 
 from pytorch_transformers.modeling_bert import (BertConfig, BertEmbeddings,
                                                 BertLayerNorm, BertModel,
@@ -144,7 +144,6 @@ class RobertaLMHead(nn.Module):
         return x
 
 
-
 class RobertaForSequenceClassification(BertPreTrainedModel):
     """
     Roberta Model with a classifier head on top.