Fix #17893, removed dead code (#17917)

* Removed dead position_id code, fix #17893

* Removed unused var

* Now ignores removed (dead) dict key for backward comp

src/transformers/models/longformer/modeling_longformer.py

@@ -447,8 +447,6 @@ class LongformerEmbeddings(nn.Module):
         self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
         self.dropout = nn.Dropout(config.hidden_dropout_prob)
 
-        # position_ids (1, len position emb) is contiguous in memory and exported when serialized
-        self.register_buffer("position_ids", torch.arange(config.max_position_embeddings).expand((1, -1)))
         self.position_embedding_type = getattr(config, "position_embedding_type", "absolute")
 
         self.padding_idx = config.pad_token_id
@@ -469,13 +467,8 @@ class LongformerEmbeddings(nn.Module):
         else:
             input_shape = inputs_embeds.size()[:-1]
 
-        seq_length = input_shape[1]
-
-        if position_ids is None:
-            position_ids = self.position_ids[:, :seq_length]
-
         if token_type_ids is None:
-            token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=self.position_ids.device)
+            token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=position_ids.device)
 
         if inputs_embeds is None:
             inputs_embeds = self.word_embeddings(input_ids)
@@ -1392,7 +1385,7 @@ class LongformerPreTrainedModel(PreTrainedModel):
     config_class = LongformerConfig
     base_model_prefix = "longformer"
     supports_gradient_checkpointing = True
-    _keys_to_ignore_on_load_missing = [r"position_ids"]
+    _keys_to_ignore_on_load_unexpected = [r"position_ids"]
 
     def _init_weights(self, module):
         """Initialize the weights"""