Fix beam_scores shape when token scores shape changes after `logits…

…_processor` (huggingface#25980)

src/transformers/generation/utils.py

@@ -3038,7 +3038,9 @@ def beam_search(
             )  # (batch_size * num_beams, vocab_size)
 
             next_token_scores_processed = logits_processor(input_ids, next_token_scores)
-            next_token_scores = next_token_scores_processed + beam_scores[:, None].expand_as(next_token_scores)
+            next_token_scores = next_token_scores_processed + beam_scores[:, None].expand_as(
+                next_token_scores_processed
+            )
 
             # Store scores, attentions and hidden_states when required
             if return_dict_in_generate:
@@ -3363,7 +3365,9 @@ def beam_sample(
             )  # (batch_size * num_beams, vocab_size)
 
             next_token_scores_processed = logits_processor(input_ids, next_token_scores)
-            next_token_scores = next_token_scores_processed + beam_scores[:, None].expand_as(next_token_scores)
+            next_token_scores = next_token_scores_processed + beam_scores[:, None].expand_as(
+                next_token_scores_processed
+            )
             # Note: logits warpers are intentionally applied after adding running beam scores. On some logits warpers
             # (like top_p) this is indiferent, but on others (like temperature) it is not. For reference, see
             # https://github.com/huggingface/transformers/pull/5420#discussion_r449779867
@@ -4080,7 +4084,9 @@ def constrained_beam_search(
 
             next_token_scores_processed = logits_processor(input_ids, next_token_scores)
 
-            next_token_scores = next_token_scores_processed + beam_scores[:, None].expand_as(next_token_scores)
+            next_token_scores = next_token_scores_processed + beam_scores[:, None].expand_as(
+                next_token_scores_processed
+            )
 
             scores_for_all_vocab = next_token_scores.clone()