Renaming ClfHead in MultipleChoiceHead.

model_pytorch.py

@@ -187,11 +187,11 @@ class LMHead(nn.Module):
         return lm_logits
 
 
-class ClfHead(nn.Module):
+class MultipleChoiceHead(nn.Module):
     """ Classifier Head for the transformer """
 
     def __init__(self, clf_token, cfg):
-        super(ClfHead, self).__init__()
+        super(MultipleChoiceHead, self).__init__()
         self.n_embd = cfg.n_embd
         self.clf_token = clf_token
         self.dropout = nn.Dropout2d(cfg.clf_pdrop)  # To reproduce the noise_shape parameter of TF implementation
@@ -216,18 +216,18 @@ class ClfHead(nn.Module):
 
 
 class DoubleHeadModel(nn.Module):
-    """ Transformer with language model and classification heads """
+    """ Transformer with language model and multiple choice heads """
     def __init__(self, cfg, clf_token, vocab=40990, n_ctx=512):
         super(DoubleHeadModel, self).__init__()
         self.transformer = TransformerModel(cfg, vocab=vocab, n_ctx=n_ctx)
         self.lm_head = LMHead(self.transformer, cfg)
-        self.clf_head = ClfHead(clf_token, cfg)
+        self.choice_head = MultipleChoiceHead(clf_token, cfg)
 
     def forward(self, x):
         h = self.transformer(x)
         lm_logits = self.lm_head(h)
-        clf_logits = self.clf_head(h, x)
-        return lm_logits, clf_logits
+        choice_logits = self.choice_head(h, x)
+        return lm_logits, choice_logits
 
 
 def load_openai_pretrained_model(model, n_ctx=-1, n_special=-1, n_transfer=12, n_embd=768, path='./model/',