Remove unnecessary views of position_ids (#26059)

* Remove unnecessary `view` of `position_ids` in `modeling_llama`

When `position_ids` is `None`, its value is generated using
`torch.arange`, which creates a tensor of size `(seq_length +
past_key_values_length) - past_key_values_length = seq_length`. The
tensor is then unsqueezed, resulting in a tensor of shape `(1,
seq_length)`. This means that the last `view` to a tensor of shape
`(-1, seq_length)` is a no-op.

This commit removes the unnecessary view.

* Remove no-op `view` of `position_ids` in rest of transformer models

src/transformers/models/codegen/modeling_codegen.py

@@ -475,9 +475,6 @@ class CodeGenModel(CodeGenPreTrainedModel):
         if token_type_ids is not None:
             token_type_ids = token_type_ids.view(-1, input_shape[-1])
 
-        if position_ids is not None:
-            position_ids = position_ids.view(-1, input_shape[-1]).long()
-
         if past_key_values is None:
             past_length = 0
             past_key_values = tuple([None] * len(self.h))
@@ -486,7 +483,7 @@ class CodeGenModel(CodeGenPreTrainedModel):
 
         if position_ids is None:
             position_ids = torch.arange(past_length, input_shape[-1] + past_length, dtype=torch.long, device=device)
-            position_ids = position_ids.unsqueeze(0).view(-1, input_shape[-1])
+            position_ids = position_ids.unsqueeze(0)
 
         # Attention mask.
         if attention_mask is not None:

src/transformers/models/ctrl/modeling_ctrl.py

@@ -416,7 +416,7 @@ class CTRLModel(CTRLPreTrainedModel):
             past_length = past_key_values[0][0].size(-2)
         if position_ids is None:
             position_ids = torch.arange(past_length, input_shape[-1] + past_length, dtype=torch.long, device=device)
-            position_ids = position_ids.unsqueeze(0).view(-1, input_shape[-1])
+            position_ids = position_ids.unsqueeze(0)
 
         # Attention mask.
         if attention_mask is not None:
@@ -447,7 +447,6 @@ class CTRLModel(CTRLPreTrainedModel):
             token_type_embeds *= np.sqrt(self.d_model_size)
         else:
             token_type_embeds = 0
-        position_ids = position_ids.view(-1, input_shape[-1])
 
         if inputs_embeds is None:
             inputs_embeds = self.w(input_ids)

src/transformers/models/decision_transformer/modeling_decision_transformer.py

@@ -544,8 +544,6 @@ class DecisionTransformerGPT2Model(DecisionTransformerGPT2PreTrainedModel):
 
         if token_type_ids is not None:
             token_type_ids = token_type_ids.view(-1, input_shape[-1])
-        if position_ids is not None:
-            position_ids = position_ids.view(-1, input_shape[-1])
 
         if past_key_values is None:
             past_length = 0
@@ -554,7 +552,7 @@ class DecisionTransformerGPT2Model(DecisionTransformerGPT2PreTrainedModel):
             past_length = past_key_values[0][0].size(-2)
         if position_ids is None:
             position_ids = torch.arange(past_length, input_shape[-1] + past_length, dtype=torch.long, device=device)
-            position_ids = position_ids.unsqueeze(0).view(-1, input_shape[-1])
+            position_ids = position_ids.unsqueeze(0)
 
         # GPT2Attention mask.
         if attention_mask is not None:

src/transformers/models/deprecated/open_llama/modeling_open_llama.py

@@ -630,9 +630,7 @@ class OpenLlamaModel(OpenLlamaPreTrainedModel):
             position_ids = torch.arange(
                 past_key_values_length, seq_length + past_key_values_length, dtype=torch.long, device=device
             )
-            position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
+            position_ids = position_ids.unsqueeze(0)
-        else:
-            position_ids = position_ids.view(-1, seq_length).long()
 
         if inputs_embeds is None:
             inputs_embeds = self.embed_tokens(input_ids)

src/transformers/models/falcon/modeling_falcon.py

@@ -1128,9 +1128,7 @@ class FalconModel(FalconPreTrainedModel):
                 position_ids = torch.arange(
                     past_key_values_length, seq_length + past_key_values_length, dtype=torch.long, device=device
                 )
-                position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
+                position_ids = position_ids.unsqueeze(0)
-            else:
-                position_ids = position_ids.view(-1, seq_length).long()
 
         causal_mask = self._prepare_attn_mask(
             attention_mask,

src/transformers/models/gpt2/modeling_gpt2.py

@@ -790,8 +790,6 @@ class GPT2Model(GPT2PreTrainedModel):
 
         if token_type_ids is not None:
             token_type_ids = token_type_ids.view(-1, input_shape[-1])
-        if position_ids is not None:
-            position_ids = position_ids.view(-1, input_shape[-1])
 
         if past_key_values is None:
             past_length = 0
@@ -800,7 +798,7 @@ class GPT2Model(GPT2PreTrainedModel):
             past_length = past_key_values[0][0].size(-2)
         if position_ids is None:
             position_ids = torch.arange(past_length, input_shape[-1] + past_length, dtype=torch.long, device=device)
-            position_ids = position_ids.unsqueeze(0).view(-1, input_shape[-1])
+            position_ids = position_ids.unsqueeze(0)
 
         # GPT2Attention mask.
         if attention_mask is not None:

src/transformers/models/gpt_bigcode/modeling_gpt_bigcode.py

@@ -577,8 +577,6 @@ class GPTBigCodeModel(GPTBigCodePreTrainedModel):
 
         if token_type_ids is not None:
             token_type_ids = token_type_ids.view(-1, input_shape[-1])
-        if position_ids is not None:
-            position_ids = position_ids.view(-1, input_shape[-1])
 
         if past_key_values is None:
             past_length = 0
@@ -594,7 +592,7 @@ class GPTBigCodeModel(GPTBigCodePreTrainedModel):
                 position_ids = position_ids[:, past_length : input_shape[-1] + past_length :]
         elif position_ids is None:
             position_ids = torch.arange(past_length, input_shape[-1] + past_length, dtype=torch.long, device=device)
-            position_ids = position_ids.unsqueeze(0).view(-1, input_shape[-1])
+            position_ids = position_ids.unsqueeze(0)
 
         # Self-attention mask.
         query_length = input_shape[-1]

src/transformers/models/gpt_neo/modeling_gpt_neo.py

@@ -539,8 +539,6 @@ class GPTNeoModel(GPTNeoPreTrainedModel):
 
         if token_type_ids is not None:
             token_type_ids = token_type_ids.view(-1, input_shape[-1])
-        if position_ids is not None:
-            position_ids = position_ids.view(-1, input_shape[-1])
 
         if past_key_values is None:
             past_length = 0
@@ -550,7 +548,7 @@ class GPTNeoModel(GPTNeoPreTrainedModel):
 
         if position_ids is None:
             position_ids = torch.arange(past_length, input_shape[-1] + past_length, dtype=torch.long, device=device)
-            position_ids = position_ids.unsqueeze(0).view(-1, input_shape[-1])
+            position_ids = position_ids.unsqueeze(0)
 
         # Attention mask.
         if attention_mask is not None:

src/transformers/models/gpt_neox/modeling_gpt_neox.py

@@ -596,9 +596,7 @@ class GPTNeoXModel(GPTNeoXPreTrainedModel):
         if position_ids is None:
             device = input_ids.device if input_ids is not None else inputs_embeds.device
             position_ids = torch.arange(past_length, seq_length + past_length, dtype=torch.long, device=device)
-            position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
+            position_ids = position_ids.unsqueeze(0)
-        else:
-            position_ids = position_ids.view(-1, seq_length).long()
 
         # Attention mask.
         if attention_mask is not None:

src/transformers/models/gptj/modeling_gptj.py

@@ -592,9 +592,6 @@ class GPTJModel(GPTJPreTrainedModel):
         if token_type_ids is not None:
             token_type_ids = token_type_ids.view(-1, input_shape[-1])
 
-        if position_ids is not None:
-            position_ids = position_ids.view(-1, input_shape[-1]).long()
-
         if past_key_values is None:
             past_length = 0
             past_key_values = tuple([None] * len(self.h))
@@ -603,7 +600,7 @@ class GPTJModel(GPTJPreTrainedModel):
 
         if position_ids is None:
             position_ids = torch.arange(past_length, input_shape[-1] + past_length, dtype=torch.long, device=device)
-            position_ids = position_ids.unsqueeze(0).view(-1, input_shape[-1])
+            position_ids = position_ids.unsqueeze(0)
 
         # Attention mask.
         if attention_mask is not None:

src/transformers/models/idefics/modeling_idefics.py

@@ -1208,9 +1208,7 @@ class IdeficsModel(IdeficsPreTrainedModel):
             position_ids = torch.arange(
                 past_key_values_length, seq_length + past_key_values_length, dtype=torch.long, device=device
             )
-            position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
+            position_ids = position_ids.unsqueeze(0)
-        else:
-            position_ids = position_ids.view(-1, seq_length).long()
 
         no_images = False
         if (pixel_values, image_encoder_embeddings, perceiver_embeddings).count(None) != 2:

src/transformers/models/imagegpt/modeling_imagegpt.py

@@ -729,8 +729,6 @@ class ImageGPTModel(ImageGPTPreTrainedModel):
 
         if token_type_ids is not None:
             token_type_ids = token_type_ids.view(-1, input_shape[-1])
-        if position_ids is not None:
-            position_ids = position_ids.view(-1, input_shape[-1])
 
         if past_key_values is None:
             past_length = 0
@@ -739,7 +737,7 @@ class ImageGPTModel(ImageGPTPreTrainedModel):
             past_length = past_key_values[0][0].size(-2)
         if position_ids is None:
             position_ids = torch.arange(past_length, input_shape[-1] + past_length, dtype=torch.long, device=device)
-            position_ids = position_ids.unsqueeze(0).view(-1, input_shape[-1])
+            position_ids = position_ids.unsqueeze(0)
 
         # ImageGPTAttention mask.
         if attention_mask is not None:

src/transformers/models/llama/modeling_llama.py

@@ -867,9 +867,7 @@ class LlamaModel(LlamaPreTrainedModel):
             position_ids = torch.arange(
                 past_key_values_length, seq_length + past_key_values_length, dtype=torch.long, device=device
             )
-            position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
+            position_ids = position_ids.unsqueeze(0)
-        else:
-            position_ids = position_ids.view(-1, seq_length).long()
 
         if inputs_embeds is None:
             inputs_embeds = self.embed_tokens(input_ids)

src/transformers/models/persimmon/modeling_persimmon.py

@@ -638,9 +638,7 @@ class PersimmonModel(PersimmonPreTrainedModel):
             position_ids = torch.arange(
                 past_key_values_length, seq_length + past_key_values_length, dtype=torch.long, device=device
             )
-            position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
+            position_ids = position_ids.unsqueeze(0)
-        else:
-            position_ids = position_ids.view(-1, seq_length).long()
 
         if inputs_embeds is None:
             inputs_embeds = self.embed_tokens(input_ids)

src/transformers/models/xglm/modeling_xglm.py

@@ -623,9 +623,7 @@ class XGLMModel(XGLMPreTrainedModel):
                 dtype=torch.long,
                 device=input_ids.device if input_ids is not None else inputs_embeds.device,
             )
-            position_ids = position_ids.unsqueeze(0).view(-1, input_shape[-1])
+            position_ids = position_ids.unsqueeze(0)
-        else:
-            position_ids = position_ids.view(-1, input_shape[-1])
 
         if inputs_embeds is None:
             inputs_embeds = self.embed_tokens(input_ids) * self.embed_scale