Remove hard-coded pad token id in distilbert and albert (#3965)

src/transformers/modeling_albert.py

@@ -175,7 +175,7 @@ class AlbertEmbeddings(BertEmbeddings):
     def __init__(self, config):
         super().__init__(config)
 
-        self.word_embeddings = nn.Embedding(config.vocab_size, config.embedding_size, padding_idx=0)
+        self.word_embeddings = nn.Embedding(config.vocab_size, config.embedding_size, padding_idx=config.pad_token_id)
         self.position_embeddings = nn.Embedding(config.max_position_embeddings, config.embedding_size)
         self.token_type_embeddings = nn.Embedding(config.type_vocab_size, config.embedding_size)
         self.LayerNorm = torch.nn.LayerNorm(config.embedding_size, eps=config.layer_norm_eps)

src/transformers/modeling_distilbert.py

@@ -61,7 +61,7 @@ def create_sinusoidal_embeddings(n_pos, dim, out):
 class Embeddings(nn.Module):
     def __init__(self, config):
         super().__init__()
-        self.word_embeddings = nn.Embedding(config.vocab_size, config.dim, padding_idx=0)
+        self.word_embeddings = nn.Embedding(config.vocab_size, config.dim, padding_idx=config.pad_token_id)
         self.position_embeddings = nn.Embedding(config.max_position_embeddings, config.dim)
         if config.sinusoidal_pos_embds:
             create_sinusoidal_embeddings(