Add support for beam search's num_return_sequencs flag in flax (#23082)

* add code for numReturnSeq

* add flax support for num return sequences

* Make Fix up for changes

* add test for num return sequences

* lint

src/transformers/generation/flax_utils.py

@@ -463,6 +463,7 @@ class FlaxGenerationMixin:
                 logits_processor=logits_processor,
                 trace=trace,
                 params=params,
+                num_return_sequences=generation_config.num_return_sequences,
                 model_kwargs=model_kwargs,
             )
         else:
@@ -749,6 +750,7 @@ class FlaxGenerationMixin:
         logits_processor: Optional[FlaxLogitsProcessorList] = None,
         trace: bool = True,
         params: Optional[Dict[str, jnp.ndarray]] = None,
+        num_return_sequences: Optional[int] = None,
         model_kwargs: Optional[Dict[str, jnp.ndarray]] = None,
     ):
         """
@@ -793,6 +795,9 @@ class FlaxGenerationMixin:
         eos_token_id = eos_token_id if eos_token_id is not None else self.generation_config.eos_token_id
         length_penalty = length_penalty if length_penalty is not None else self.generation_config.length_penalty
         early_stopping = early_stopping if early_stopping is not None else self.generation_config.early_stopping
+        num_return_sequences = (
+            num_return_sequences if num_return_sequences is not None else self.generation_config.num_return_sequences
+        )
 
         batch_size, num_beams, cur_len = input_ids.shape
 
@@ -996,8 +1001,8 @@ class FlaxGenerationMixin:
         sequences = jnp.where(none_finished[:, None, None], state.sequences, state.running_sequences)
         scores = jnp.where(none_finished[:, None], state.scores, state.running_scores)
 
-        # take best beam for each batch
+        # Take best beams for each batch (the score is sorted in descending order)
-        sequences = sequences[:, 0]
+        sequences = flatten_beam_dim(sequences[:, :num_return_sequences, :])
-        scores = scores[:, 0]
+        scores = flatten_beam_dim(scores[:, :num_return_sequences])
 
         return FlaxBeamSearchOutput(sequences=sequences, scores=scores)

tests/generation/test_flax_utils.py

@@ -158,6 +158,19 @@ class FlaxGenerationTesterMixin:
 
             self.assertListEqual(generation_outputs.tolist(), jit_generation_outputs.tolist())
 
+    def test_beam_search_generate_num_return_sequences(self):
+        config, input_ids, _, max_length = self._get_input_ids_and_config()
+        config.do_sample = False
+        config.max_length = max_length
+        config.num_beams = 2
+        config.num_return_sequences = 2
+
+        for model_class in self.all_generative_model_classes:
+            model = model_class(config)
+
+            generation_outputs = model.generate(input_ids).sequences
+            self.assertEqual(generation_outputs.shape[0], input_ids.shape[0] * config.num_return_sequences)
+
     def test_sample_generate_logits_warper(self):
         config, input_ids, _, max_length = self._get_input_ids_and_config()
         config.do_sample = True