Remove "double" assignment in TF-BART like models (#9997)

* Replace `attn_weights = attn_wegihts = tf.reshape(...)`
with `attn_weights = tf.reshape(...)` and thus remove
unintentionally used "double" assignment.

src/transformers/models/bart/modeling_tf_bart.py

@@ -241,7 +241,7 @@ class TFBartAttention(tf.keras.layers.Layer):
             attn_weights = tf.reshape(layer_head_mask, (1, -1, 1, 1)) * tf.reshape(
                 attn_weights, (bsz, self.num_heads, tgt_len, src_len)
             )
-            attn_weights = attn_weights = tf.reshape(attn_weights, (bsz * self.num_heads, tgt_len, src_len))
+            attn_weights = tf.reshape(attn_weights, (bsz * self.num_heads, tgt_len, src_len))
 
         attn_probs = self.dropout(attn_weights, training=training)
 

src/transformers/models/blenderbot/modeling_tf_blenderbot.py

@@ -244,7 +244,7 @@ class TFBlenderbotAttention(tf.keras.layers.Layer):
             attn_weights = tf.reshape(layer_head_mask, (1, -1, 1, 1)) * tf.reshape(
                 attn_weights, (bsz, self.num_heads, tgt_len, src_len)
             )
-            attn_weights = attn_weights = tf.reshape(attn_weights, (bsz * self.num_heads, tgt_len, src_len))
+            attn_weights = tf.reshape(attn_weights, (bsz * self.num_heads, tgt_len, src_len))
 
         attn_probs = self.dropout(attn_weights, training=training)
 

src/transformers/models/blenderbot_small/modeling_tf_blenderbot_small.py

@@ -243,7 +243,7 @@ class TFBlenderbotSmallAttention(tf.keras.layers.Layer):
             attn_weights = tf.reshape(layer_head_mask, (1, -1, 1, 1)) * tf.reshape(
                 attn_weights, (bsz, self.num_heads, tgt_len, src_len)
             )
-            attn_weights = attn_weights = tf.reshape(attn_weights, (bsz * self.num_heads, tgt_len, src_len))
+            attn_weights = tf.reshape(attn_weights, (bsz * self.num_heads, tgt_len, src_len))
 
         attn_probs = self.dropout(attn_weights, training=training)
 

src/transformers/models/marian/modeling_tf_marian.py

@@ -273,7 +273,7 @@ class TFMarianAttention(tf.keras.layers.Layer):
             attn_weights = tf.reshape(layer_head_mask, (1, -1, 1, 1)) * tf.reshape(
                 attn_weights, (bsz, self.num_heads, tgt_len, src_len)
             )
-            attn_weights = attn_weights = tf.reshape(attn_weights, (bsz * self.num_heads, tgt_len, src_len))
+            attn_weights = tf.reshape(attn_weights, (bsz * self.num_heads, tgt_len, src_len))
 
         attn_probs = self.dropout(attn_weights, training=training)
 

src/transformers/models/mbart/modeling_tf_mbart.py

@@ -247,7 +247,7 @@ class TFMBartAttention(tf.keras.layers.Layer):
             attn_weights = tf.reshape(layer_head_mask, (1, -1, 1, 1)) * tf.reshape(
                 attn_weights, (bsz, self.num_heads, tgt_len, src_len)
             )
-            attn_weights = attn_weights = tf.reshape(attn_weights, (bsz * self.num_heads, tgt_len, src_len))
+            attn_weights = tf.reshape(attn_weights, (bsz * self.num_heads, tgt_len, src_len))
 
         attn_probs = self.dropout(attn_weights, training=training)
 

src/transformers/models/pegasus/modeling_tf_pegasus.py

@@ -274,7 +274,7 @@ class TFPegasusAttention(tf.keras.layers.Layer):
             attn_weights = tf.reshape(layer_head_mask, (1, -1, 1, 1)) * tf.reshape(
                 attn_weights, (bsz, self.num_heads, tgt_len, src_len)
             )
-            attn_weights = attn_weights = tf.reshape(attn_weights, (bsz * self.num_heads, tgt_len, src_len))
+            attn_weights = tf.reshape(attn_weights, (bsz * self.num_heads, tgt_len, src_len))
 
         attn_probs = self.dropout(attn_weights, training=training)