Add warning for missing attention mask when pad tokens are detected (#25345)

* Add attention mask and pad token warning to many of the models

* Remove changes under examples/research_projects

These files are not maintained by HG.

* Skip the warning check during torch.fx or JIT tracing

* Switch ordering for the warning and input shape assignment

This ordering is a little cleaner for some of the cases.

* Add missing line break in one of the files

src/transformers/models/deberta/modeling_deberta.py

@@ -959,6 +959,7 @@ class DebertaModel(DebertaPreTrainedModel):
         if input_ids is not None and inputs_embeds is not None:
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
         elif input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
             input_shape = input_ids.size()
         elif inputs_embeds is not None:
             input_shape = inputs_embeds.size()[:-1]

src/transformers/models/deberta_v2/modeling_deberta_v2.py

@@ -1059,6 +1059,7 @@ class DebertaV2Model(DebertaV2PreTrainedModel):
         if input_ids is not None and inputs_embeds is not None:
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
         elif input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
             input_shape = input_ids.size()
         elif inputs_embeds is not None:
             input_shape = inputs_embeds.size()[:-1]

src/transformers/models/decision_transformer/modeling_decision_transformer.py

@@ -530,6 +530,7 @@ class DecisionTransformerGPT2Model(DecisionTransformerGPT2PreTrainedModel):
         if input_ids is not None and inputs_embeds is not None:
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
         elif input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
             input_shape = input_ids.size()
             input_ids = input_ids.view(-1, input_shape[-1])
             batch_size = input_ids.shape[0]

src/transformers/models/distilbert/modeling_distilbert.py

@@ -589,6 +589,7 @@ class DistilBertModel(DistilBertPreTrainedModel):
         if input_ids is not None and inputs_embeds is not None:
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
         elif input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
             input_shape = input_ids.size()
         elif inputs_embeds is not None:
             input_shape = inputs_embeds.size()[:-1]

src/transformers/models/dpr/modeling_dpr.py

@@ -562,6 +562,7 @@ class DPRQuestionEncoder(DPRPretrainedQuestionEncoder):
         if input_ids is not None and inputs_embeds is not None:
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
         elif input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
             input_shape = input_ids.size()
         elif inputs_embeds is not None:
             input_shape = inputs_embeds.size()[:-1]
@@ -650,6 +651,7 @@ class DPRReader(DPRPretrainedReader):
         if input_ids is not None and inputs_embeds is not None:
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
         elif input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
             input_shape = input_ids.size()
         elif inputs_embeds is not None:
             input_shape = inputs_embeds.size()[:-1]

src/transformers/models/electra/modeling_electra.py

@@ -868,6 +868,7 @@ class ElectraModel(ElectraPreTrainedModel):
         if input_ids is not None and inputs_embeds is not None:
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
         elif input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
             input_shape = input_ids.size()
         elif inputs_embeds is not None:
             input_shape = inputs_embeds.size()[:-1]

src/transformers/models/ernie/modeling_ernie.py

@@ -895,6 +895,7 @@ class ErnieModel(ErniePreTrainedModel):
         if input_ids is not None and inputs_embeds is not None:
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
         elif input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
             input_shape = input_ids.size()
         elif inputs_embeds is not None:
             input_shape = inputs_embeds.size()[:-1]

src/transformers/models/esm/modeling_esm.py

@@ -876,6 +876,7 @@ class EsmModel(EsmPreTrainedModel):
         if input_ids is not None and inputs_embeds is not None:
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
         elif input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
             input_shape = input_ids.size()
         elif inputs_embeds is not None:
             input_shape = inputs_embeds.size()[:-1]

src/transformers/models/funnel/modeling_funnel.py

@@ -966,6 +966,7 @@ class FunnelBaseModel(FunnelPreTrainedModel):
         if input_ids is not None and inputs_embeds is not None:
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
         elif input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
             input_shape = input_ids.size()
         elif inputs_embeds is not None:
             input_shape = inputs_embeds.size()[:-1]
@@ -1041,6 +1042,7 @@ class FunnelModel(FunnelPreTrainedModel):
         if input_ids is not None and inputs_embeds is not None:
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
         elif input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
             input_shape = input_ids.size()
         elif inputs_embeds is not None:
             input_shape = inputs_embeds.size()[:-1]

src/transformers/models/git/modeling_git.py

@@ -1213,6 +1213,7 @@ class GitModel(GitPreTrainedModel):
         if input_ids is not None and inputs_embeds is not None:
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
         elif input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
             input_shape = input_ids.size()
         elif inputs_embeds is not None:
             input_shape = inputs_embeds.size()[:-1]

src/transformers/models/gpt2/modeling_gpt2.py

@@ -776,6 +776,7 @@ class GPT2Model(GPT2PreTrainedModel):
         if input_ids is not None and inputs_embeds is not None:
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
         elif input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
             input_shape = input_ids.size()
             input_ids = input_ids.view(-1, input_shape[-1])
             batch_size = input_ids.shape[0]

src/transformers/models/gpt_bigcode/modeling_gpt_bigcode.py

@@ -560,6 +560,7 @@ class GPTBigCodeModel(GPTBigCodePreTrainedModel):
         if input_ids is not None and inputs_embeds is not None:
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
         elif input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
             input_shape = input_ids.size()
             input_ids = input_ids.view(-1, input_shape[-1])
             batch_size = input_ids.shape[0]

src/transformers/models/gpt_neo/modeling_gpt_neo.py

@@ -525,6 +525,7 @@ class GPTNeoModel(GPTNeoPreTrainedModel):
         if input_ids is not None and inputs_embeds is not None:
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
         elif input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
             input_shape = input_ids.size()
             input_ids = input_ids.view(-1, input_shape[-1])
             batch_size = input_ids.shape[0]

src/transformers/models/gpt_neox/modeling_gpt_neox.py

@@ -571,6 +571,7 @@ class GPTNeoXModel(GPTNeoXPreTrainedModel):
         if input_ids is not None and inputs_embeds is not None:
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
         elif input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
             input_shape = input_ids.size()
         elif inputs_embeds is not None:
             input_shape = inputs_embeds.size()[:-1]

src/transformers/models/gpt_neox_japanese/modeling_gpt_neox_japanese.py

@@ -508,6 +508,7 @@ class GPTNeoXJapaneseModel(GPTNeoXJapanesePreTrainedModel):
         if input_ids is not None and inputs_embeds is not None:
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
         elif input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
             input_shape = input_ids.size()
         elif inputs_embeds is not None:
             input_shape = inputs_embeds.size()[:-1]

src/transformers/models/gptj/modeling_gptj.py

@@ -577,6 +577,7 @@ class GPTJModel(GPTJPreTrainedModel):
         if input_ids is not None and inputs_embeds is not None:
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
         elif input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
             input_shape = input_ids.size()
             input_ids = input_ids.view(-1, input_shape[-1])
             batch_size = input_ids.shape[0]

src/transformers/models/ibert/modeling_ibert.py

@@ -796,6 +796,7 @@ class IBertModel(IBertPreTrainedModel):
         if input_ids is not None and inputs_embeds is not None:
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
         elif input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
             input_shape = input_ids.size()
         elif inputs_embeds is not None:
             input_shape = inputs_embeds.size()[:-1]

src/transformers/models/imagegpt/modeling_imagegpt.py

@@ -715,6 +715,7 @@ class ImageGPTModel(ImageGPTPreTrainedModel):
         if input_ids is not None and inputs_embeds is not None:
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
         elif input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
             input_shape = input_ids.size()
             input_ids = input_ids.view(-1, input_shape[-1])
             batch_size = input_ids.shape[0]

src/transformers/models/layoutlm/modeling_layoutlm.py

@@ -795,6 +795,7 @@ class LayoutLMModel(LayoutLMPreTrainedModel):
         if input_ids is not None and inputs_embeds is not None:
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
         elif input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
             input_shape = input_ids.size()
         elif inputs_embeds is not None:
             input_shape = inputs_embeds.size()[:-1]

src/transformers/models/layoutlmv2/modeling_layoutlmv2.py

@@ -1047,6 +1047,7 @@ class LayoutLMv2ForSequenceClassification(LayoutLMv2PreTrainedModel):
         if input_ids is not None and inputs_embeds is not None:
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
         elif input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
             input_shape = input_ids.size()
         elif inputs_embeds is not None:
             input_shape = inputs_embeds.size()[:-1]