correct greedy generation when doing beam search (#3078)

* correct greedy generation when doing beam search

* improve comment

src/transformers/modeling_utils.py

@@ -754,6 +754,19 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
         else:
             assert input_ids.dim() == 2, "Input prompt should be of shape (batch_size, sequence length)."
 
+        if do_sample is False:
+            if num_beams == 1:
+                # no_beam_search greedy generation conditions
+                assert (
+                    num_return_sequences == 1
+                ), "Greedy decoding will always produce the same output for num_beams == 1 and num_return_sequences > 1. Please set num_return_sequences = 1"
+
+            else:
+                # beam_search greedy generation conditions
+                assert (
+                    num_beams >= num_return_sequences
+                ), "Greedy beam search decoding cannot return more sequences than it has beams. Please set num_beams >= num_return_sequences"
+
         if pad_token_id is None and eos_token_ids is not None:
             logger.warning(
                 "Setting `pad_token_id` to {} (first `eos_token_id`) to generate sequence".format(eos_token_ids[0])
@@ -764,7 +777,7 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
         cur_len = input_ids.shape[1]
         vocab_size = self.config.vocab_size
 
-        if num_return_sequences != 1:
+        if num_return_sequences != 1 and do_sample:
             # Expand input to num return sequences
             input_ids = input_ids.unsqueeze(1).expand(batch_size, num_return_sequences, cur_len)
             input_ids = input_ids.contiguous().view(
@@ -787,6 +800,7 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
                 pad_token_id,
                 eos_token_ids,
                 effective_batch_size,
+                num_return_sequences,
                 length_penalty,
                 num_beams,
                 vocab_size,
@@ -826,6 +840,7 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
             All returned sequence are generated independantly.
         """
         # current position / max lengths / length of generated sentences / unfinished sentences
+
         unfinished_sents = input_ids.new(batch_size).fill_(1)
         sent_lengths = input_ids.new(batch_size).fill_(max_length)
 
@@ -906,12 +921,14 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
         pad_token_id,
         eos_token_ids,
         batch_size,
+        num_return_sequences,
         length_penalty,
         num_beams,
         vocab_size,
     ):
         """ Generate sequences for each example with beam search.
         """
+
         # Expand input to num beams
         # assert input_ids.shape == (batch_size * num_beams, cur_len)
         input_ids = input_ids.unsqueeze(1).expand(batch_size, num_beams, cur_len)
@@ -1057,20 +1074,28 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin):
                         input_ids[batch_idx * num_beams + beam_id, :cur_len].clone(), score.item()
                     )
 
+        # depending on whether greedy generation is wanted or not define different output_batch_size and output_num_return_sequences_per_batch
+        output_batch_size = batch_size if do_sample else batch_size * num_return_sequences
+        output_num_return_sequences_per_batch = 1 if do_sample else num_return_sequences
+
         # select the best hypotheses
-        sent_lengths = input_ids.new(batch_size)
+        sent_lengths = input_ids.new(output_batch_size)
         best = []
 
+        # retrieve best hypotheses
         for i, hypotheses in enumerate(generated_hyps):
-            best_hyp = max(hypotheses.beams, key=lambda x: x[0])[1]
+            sorted_hyps = sorted(hypotheses.beams, key=lambda x: x[0])
-            sent_lengths[i] = len(best_hyp)
+            for j in range(output_num_return_sequences_per_batch):
+                effective_batch_idx = output_num_return_sequences_per_batch * i + j
+                best_hyp = sorted_hyps.pop()[1]
+                sent_lengths[effective_batch_idx] = len(best_hyp)
                 best.append(best_hyp)
 
         # shorter batches are filled with pad_token
         if sent_lengths.min().item() != sent_lengths.max().item():
             assert pad_token_id is not None, "`Pad_token_id` has to be defined"
             sent_max_len = min(sent_lengths.max().item() + 1, max_length)
-            decoded = input_ids.new(batch_size, sent_max_len).fill_(pad_token_id)
+            decoded = input_ids.new(output_batch_size, sent_max_len).fill_(pad_token_id)
 
             # fill with hypothesis and eos_token_id if necessary
             for i, hypo in enumerate(best):

tests/test_modeling_common.py

@@ -621,10 +621,19 @@ class ModelTesterMixin:
                 # batch_size = 1, num_beams > 1
                 self._check_generated_tokens(model.generate(max_length=5, num_beams=3))
 
+            with self.assertRaises(AssertionError):
+                # generating multiple sequences when greedy no beam generation
+                # is not allowed as it would always generate the same sequences
+                model.generate(input_ids, do_sample=False, num_return_sequences=2)
+
+            with self.assertRaises(AssertionError):
+                # generating more sequences than having beams leads is not possible
+                model.generate(input_ids, do_sample=False, num_return_sequences=3, num_beams=2)
+
             # batch_size > 1, sample
             self._check_generated_tokens(model.generate(input_ids, num_return_sequences=3))
             # batch_size > 1, greedy
-            self._check_generated_tokens(model.generate(input_ids, do_sample=False, num_return_sequences=3))
+            self._check_generated_tokens(model.generate(input_ids, do_sample=False))
             # batch_size > 1, num_beams > 1, sample
             self._check_generated_tokens(model.generate(input_ids, num_beams=3, num_return_sequences=3,))
             # batch_size > 1, num_beams > 1, greedy