Change the way tensor is reshaped in BartAttention (from .view to .reshape) (#21860)

* Change the .view call to .reshape

* Change the .view call to .reshape to all the copies from bart attention

* Fix copies and style

* Fix copies and style

* Fix copies and style

* Fix copies and style

* Fix copies and style

* Revert unneccessary changes

* Revert unneccessary changes

* Revert unneccessary changes

* Revert unneccessary changes

src/transformers/models/bart/modeling_bart.py

@@ -229,8 +229,8 @@ class BartAttention(nn.Module):
 
         proj_shape = (bsz * self.num_heads, -1, self.head_dim)
         query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
-        key_states = key_states.view(*proj_shape)
-        value_states = value_states.view(*proj_shape)
+        key_states = key_states.reshape(*proj_shape)
+        value_states = value_states.reshape(*proj_shape)
 
         src_len = key_states.size(1)
         attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))

src/transformers/models/bigbird_pegasus/modeling_bigbird_pegasus.py

@@ -1288,8 +1288,8 @@ class BigBirdPegasusDecoderAttention(nn.Module):
 
         proj_shape = (bsz * self.num_heads, -1, self.head_dim)
         query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
-        key_states = key_states.view(*proj_shape)
-        value_states = value_states.view(*proj_shape)
+        key_states = key_states.reshape(*proj_shape)
+        value_states = value_states.reshape(*proj_shape)
 
         src_len = key_states.size(1)
         attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))

src/transformers/models/biogpt/modeling_biogpt.py

@@ -190,8 +190,8 @@ class BioGptAttention(nn.Module):
 
         proj_shape = (bsz * self.num_heads, -1, self.head_dim)
         query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
-        key_states = key_states.view(*proj_shape)
-        value_states = value_states.view(*proj_shape)
+        key_states = key_states.reshape(*proj_shape)
+        value_states = value_states.reshape(*proj_shape)
 
         src_len = key_states.size(1)
         attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))

src/transformers/models/blenderbot/modeling_blenderbot.py

@@ -216,8 +216,8 @@ class BlenderbotAttention(nn.Module):
 
         proj_shape = (bsz * self.num_heads, -1, self.head_dim)
         query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
-        key_states = key_states.view(*proj_shape)
-        value_states = value_states.view(*proj_shape)
+        key_states = key_states.reshape(*proj_shape)
+        value_states = value_states.reshape(*proj_shape)
 
         src_len = key_states.size(1)
         attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))

src/transformers/models/blenderbot_small/modeling_blenderbot_small.py

@@ -213,8 +213,8 @@ class BlenderbotSmallAttention(nn.Module):
 
         proj_shape = (bsz * self.num_heads, -1, self.head_dim)
         query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
-        key_states = key_states.view(*proj_shape)
-        value_states = value_states.view(*proj_shape)
+        key_states = key_states.reshape(*proj_shape)
+        value_states = value_states.reshape(*proj_shape)
 
         src_len = key_states.size(1)
         attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))

src/transformers/models/data2vec/modeling_data2vec_audio.py

@@ -419,8 +419,8 @@ class Data2VecAudioAttention(nn.Module):
 
         proj_shape = (bsz * self.num_heads, -1, self.head_dim)
         query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
-        key_states = key_states.view(*proj_shape)
-        value_states = value_states.view(*proj_shape)
+        key_states = key_states.reshape(*proj_shape)
+        value_states = value_states.reshape(*proj_shape)
 
         src_len = key_states.size(1)
         attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))

src/transformers/models/gptsan_japanese/modeling_gptsan_japanese.py

@@ -451,8 +451,8 @@ class GPTSanJapaneseAttention(nn.Module):
 
         proj_shape = (bsz * self.num_heads, -1, self.head_dim)
         query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
-        key_states = key_states.view(*proj_shape)
-        value_states = value_states.view(*proj_shape)
+        key_states = key_states.reshape(*proj_shape)
+        value_states = value_states.reshape(*proj_shape)
 
         src_len = key_states.size(1)
         attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))

src/transformers/models/hubert/modeling_hubert.py

@@ -482,8 +482,8 @@ class HubertAttention(nn.Module):
 
         proj_shape = (bsz * self.num_heads, -1, self.head_dim)
         query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
-        key_states = key_states.view(*proj_shape)
-        value_states = value_states.view(*proj_shape)
+        key_states = key_states.reshape(*proj_shape)
+        value_states = value_states.reshape(*proj_shape)
 
         src_len = key_states.size(1)
         attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))

src/transformers/models/m2m_100/modeling_m2m_100.py

@@ -284,8 +284,8 @@ class M2M100Attention(nn.Module):
 
         proj_shape = (bsz * self.num_heads, -1, self.head_dim)
         query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
-        key_states = key_states.view(*proj_shape)
-        value_states = value_states.view(*proj_shape)
+        key_states = key_states.reshape(*proj_shape)
+        value_states = value_states.reshape(*proj_shape)
 
         src_len = key_states.size(1)
         attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))

src/transformers/models/marian/modeling_marian.py

@@ -231,8 +231,8 @@ class MarianAttention(nn.Module):
 
         proj_shape = (bsz * self.num_heads, -1, self.head_dim)
         query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
-        key_states = key_states.view(*proj_shape)
-        value_states = value_states.view(*proj_shape)
+        key_states = key_states.reshape(*proj_shape)
+        value_states = value_states.reshape(*proj_shape)
 
         src_len = key_states.size(1)
         attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))

src/transformers/models/mbart/modeling_mbart.py

@@ -225,8 +225,8 @@ class MBartAttention(nn.Module):
 
         proj_shape = (bsz * self.num_heads, -1, self.head_dim)
         query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
-        key_states = key_states.view(*proj_shape)
-        value_states = value_states.view(*proj_shape)
+        key_states = key_states.reshape(*proj_shape)
+        value_states = value_states.reshape(*proj_shape)
 
         src_len = key_states.size(1)
         attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))

src/transformers/models/pegasus/modeling_pegasus.py

@@ -231,8 +231,8 @@ class PegasusAttention(nn.Module):
 
         proj_shape = (bsz * self.num_heads, -1, self.head_dim)
         query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
-        key_states = key_states.view(*proj_shape)
-        value_states = value_states.view(*proj_shape)
+        key_states = key_states.reshape(*proj_shape)
+        value_states = value_states.reshape(*proj_shape)
 
         src_len = key_states.size(1)
         attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))

src/transformers/models/pegasus_x/modeling_pegasus_x.py

@@ -240,8 +240,8 @@ class PegasusXAttention(nn.Module):
 
         proj_shape = (bsz * self.num_heads, -1, self.head_dim)
         query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
-        key_states = key_states.view(*proj_shape)
-        value_states = value_states.view(*proj_shape)
+        key_states = key_states.reshape(*proj_shape)
+        value_states = value_states.reshape(*proj_shape)
 
         src_len = key_states.size(1)
         attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))

src/transformers/models/plbart/modeling_plbart.py

@@ -224,8 +224,8 @@ class PLBartAttention(nn.Module):
 
         proj_shape = (bsz * self.num_heads, -1, self.head_dim)
         query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
-        key_states = key_states.view(*proj_shape)
-        value_states = value_states.view(*proj_shape)
+        key_states = key_states.reshape(*proj_shape)
+        value_states = value_states.reshape(*proj_shape)
 
         src_len = key_states.size(1)
         attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))

src/transformers/models/sew/modeling_sew.py

@@ -482,8 +482,8 @@ class SEWAttention(nn.Module):
 
         proj_shape = (bsz * self.num_heads, -1, self.head_dim)
         query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
-        key_states = key_states.view(*proj_shape)
-        value_states = value_states.view(*proj_shape)
+        key_states = key_states.reshape(*proj_shape)
+        value_states = value_states.reshape(*proj_shape)
 
         src_len = key_states.size(1)
         attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))

src/transformers/models/speech_to_text/modeling_speech_to_text.py

@@ -291,8 +291,8 @@ class Speech2TextAttention(nn.Module):
 
         proj_shape = (bsz * self.num_heads, -1, self.head_dim)
         query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
-        key_states = key_states.view(*proj_shape)
-        value_states = value_states.view(*proj_shape)
+        key_states = key_states.reshape(*proj_shape)
+        value_states = value_states.reshape(*proj_shape)
 
         src_len = key_states.size(1)
         attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))

src/transformers/models/speech_to_text_2/modeling_speech_to_text_2.py

@@ -237,8 +237,8 @@ class Speech2Text2Attention(nn.Module):
 
         proj_shape = (bsz * self.num_heads, -1, self.head_dim)
         query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
-        key_states = key_states.view(*proj_shape)
-        value_states = value_states.view(*proj_shape)
+        key_states = key_states.reshape(*proj_shape)
+        value_states = value_states.reshape(*proj_shape)
 
         src_len = key_states.size(1)
         attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))

src/transformers/models/time_series_transformer/modeling_time_series_transformer.py

@@ -717,8 +717,8 @@ class TimeSeriesTransformerAttention(nn.Module):
 
         proj_shape = (bsz * self.num_heads, -1, self.head_dim)
         query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
-        key_states = key_states.view(*proj_shape)
-        value_states = value_states.view(*proj_shape)
+        key_states = key_states.reshape(*proj_shape)
+        value_states = value_states.reshape(*proj_shape)
 
         src_len = key_states.size(1)
         attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))

src/transformers/models/unispeech/modeling_unispeech.py

@@ -517,8 +517,8 @@ class UniSpeechAttention(nn.Module):
 
         proj_shape = (bsz * self.num_heads, -1, self.head_dim)
         query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
-        key_states = key_states.view(*proj_shape)
-        value_states = value_states.view(*proj_shape)
+        key_states = key_states.reshape(*proj_shape)
+        value_states = value_states.reshape(*proj_shape)
 
         src_len = key_states.size(1)
         attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))

src/transformers/models/unispeech_sat/modeling_unispeech_sat.py

@@ -531,8 +531,8 @@ class UniSpeechSatAttention(nn.Module):
 
         proj_shape = (bsz * self.num_heads, -1, self.head_dim)
         query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
-        key_states = key_states.view(*proj_shape)
-        value_states = value_states.view(*proj_shape)
+        key_states = key_states.reshape(*proj_shape)
+        value_states = value_states.reshape(*proj_shape)
 
         src_len = key_states.size(1)
         attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))

src/transformers/models/wav2vec2/modeling_wav2vec2.py

@@ -574,8 +574,8 @@ class Wav2Vec2Attention(nn.Module):
 
         proj_shape = (bsz * self.num_heads, -1, self.head_dim)
         query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
-        key_states = key_states.view(*proj_shape)
-        value_states = value_states.view(*proj_shape)
+        key_states = key_states.reshape(*proj_shape)
+        value_states = value_states.reshape(*proj_shape)
 
         src_len = key_states.size(1)
         attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))

src/transformers/models/whisper/modeling_whisper.py

@@ -319,8 +319,8 @@ class WhisperAttention(nn.Module):
 
         proj_shape = (bsz * self.num_heads, -1, self.head_dim)
         query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
-        key_states = key_states.view(*proj_shape)
-        value_states = value_states.view(*proj_shape)
+        key_states = key_states.reshape(*proj_shape)
+        value_states = value_states.reshape(*proj_shape)
 
         src_len = key_states.size(1)
         attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))