Fix for the issue of device-id getting hardcoded for token_type_ids during...

Fix for the issue of device-id getting hardcoded for token_type_ids during Tracing for ConvBert (#12287)

* added token_type_ids buffer to fix the issue #5664

* Handling the case that position_id buffer is not registered

* added token_type_ids buffer to fix the issue #5664

* modified to support device conversion when the model is traced

src/transformers/models/convbert/modeling_convbert.py

@@ -21,6 +21,7 @@ from operator import attrgetter
 
 import torch
 import torch.utils.checkpoint
+from packaging import version
 from torch import nn
 from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
 
@@ -196,9 +197,14 @@ class ConvBertEmbeddings(nn.Module):
         # any TensorFlow checkpoint file
         self.LayerNorm = nn.LayerNorm(config.embedding_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)))
+        if version.parse(torch.__version__) > version.parse("1.6.0"):
+            self.register_buffer(
+                "token_type_ids",
+                torch.zeros(self.position_ids.size(), dtype=torch.long, device=self.position_ids.device),
+                persistent=False,
+            )
 
     def forward(self, input_ids=None, token_type_ids=None, position_ids=None, inputs_embeds=None):
         if input_ids is not None:
@@ -211,8 +217,16 @@ class ConvBertEmbeddings(nn.Module):
         if position_ids is None:
             position_ids = self.position_ids[:, :seq_length]
 
+        # Setting the token_type_ids to the registered buffer in constructor where it is all zeros, which usually occurs
+        # when its auto-generated, registered buffer helps users when tracing the model without passing token_type_ids, solves
+        # issue #5664
         if token_type_ids is None:
-            token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=self.position_ids.device)
+            if hasattr(self, "token_type_ids"):
+                buffered_token_type_ids = self.token_type_ids[:, :seq_length]
+                buffered_token_type_ids_expanded = buffered_token_type_ids.expand(input_shape[0], seq_length)
+                token_type_ids = buffered_token_type_ids_expanded
+            else:
+                token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=self.position_ids.device)
 
         if inputs_embeds is None:
             inputs_embeds = self.word_embeddings(input_ids)
@@ -800,6 +814,7 @@ class ConvBertModel(ConvBertPreTrainedModel):
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
         elif input_ids is not None:
             input_shape = input_ids.size()
+            batch_size, seq_length = input_shape
         elif inputs_embeds is not None:
             input_shape = inputs_embeds.size()[:-1]
         else:
@@ -810,7 +825,12 @@ class ConvBertModel(ConvBertPreTrainedModel):
         if attention_mask is None:
             attention_mask = torch.ones(input_shape, device=device)
         if token_type_ids is None:
-            token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=device)
+            if hasattr(self.embeddings, "token_type_ids"):
+                buffered_token_type_ids = self.embeddings.token_type_ids[:, :seq_length]
+                buffered_token_type_ids_expanded = buffered_token_type_ids.expand(batch_size, seq_length)
+                token_type_ids = buffered_token_type_ids_expanded
+            else:
+                token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=device)
 
         extended_attention_mask = self.get_extended_attention_mask(attention_mask, input_shape, device)
         head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers)