[Generation] Fix Transition probs (huggingface#17311)

* [Draft] fix transition probs

* up

* up

* up

* make it work

* fix

* finish

* update

Author: patrickvonplaten

src/transformers/generation_beam_search.py

@@ -212,6 +212,7 @@ def process(
         next_indices: torch.LongTensor,
         pad_token_id: Optional[int] = None,
         eos_token_id: Optional[int] = None,
+        beam_indices: Optional[torch.LongTensor] = None,
     ) -> Tuple[torch.Tensor]:
         cur_len = input_ids.shape[-1]
         batch_size = len(self._beam_hyps)
@@ -256,9 +257,16 @@ def process(
                     is_beam_token_worse_than_top_num_beams = beam_token_rank >= self.group_size
                     if is_beam_token_worse_than_top_num_beams:
                         continue
+                    if beam_indices is not None:
+                        beam_index = beam_indices[batch_beam_idx]
+                        beam_index = beam_index + (next_index,)
+                    else:
+                        beam_index = None
+
                     beam_hyp.add(
                         input_ids[batch_beam_idx].clone(),
                         next_score.item(),
+                        beam_indices=beam_index,
                     )
                 else:
                     # add next predicted token since it is not eos_token
@@ -299,6 +307,7 @@ def finalize(
         max_length: int,
         pad_token_id: Optional[int] = None,
         eos_token_id: Optional[int] = None,
+        beam_indices: Optional[torch.LongTensor] = None,
     ) -> Tuple[torch.LongTensor]:
         batch_size = len(self._beam_hyps)
 
@@ -313,11 +322,13 @@ def finalize(
                 batch_beam_idx = batch_idx * self.num_beams + beam_id
                 final_score = final_beam_scores[batch_beam_idx].item()
                 final_tokens = input_ids[batch_beam_idx]
-                beam_hyp.add(final_tokens, final_score)
+                beam_index = beam_indices[batch_beam_idx] if beam_indices is not None else None
+                beam_hyp.add(final_tokens, final_score, beam_indices=beam_index)
 
         # select the best hypotheses
         sent_lengths = input_ids.new(batch_size * self.num_beam_hyps_to_keep)
         best = []
+        best_indices = []
         best_scores = torch.zeros(batch_size * self.num_beam_hyps_to_keep, device=self.device, dtype=torch.float32)
 
         # retrieve best hypotheses
@@ -327,30 +338,50 @@ def finalize(
                 best_hyp_tuple = sorted_hyps.pop()
                 best_score = best_hyp_tuple[0]
                 best_hyp = best_hyp_tuple[1]
+                best_index = best_hyp_tuple[2]
                 sent_lengths[self.num_beam_hyps_to_keep * i + j] = len(best_hyp)
 
-                # append to lists
+                # append hyp to lists
                 best.append(best_hyp)
+
+                # append indices to list
+                best_indices.append(best_index)
+
                 best_scores[i * self.num_beam_hyps_to_keep + j] = best_score
 
         # prepare for adding eos
         sent_lengths_max = sent_lengths.max().item() + 1
         sent_max_len = min(sent_lengths_max, max_length) if max_length is not None else sent_lengths_max
         decoded: torch.LongTensor = input_ids.new(batch_size * self.num_beam_hyps_to_keep, sent_max_len)
+
+        if len(best_indices) > 0 and best_indices[0] is not None:
+            indices: torch.LongTensor = input_ids.new(batch_size * self.num_beam_hyps_to_keep, sent_max_len)
+        else:
+            indices = None
+
         # shorter batches are padded if needed
         if sent_lengths.min().item() != sent_lengths.max().item():
             assert pad_token_id is not None, "`pad_token_id` has to be defined"
             decoded.fill_(pad_token_id)
+
+        if indices is not None:
+            indices.fill_(-1)
+
         # fill with hypotheses and eos_token_id if the latter fits in
-        for i, hypo in enumerate(best):
+        for i, (hypo, best_idx) in enumerate(zip(best, best_indices)):
             decoded[i, : sent_lengths[i]] = hypo
+
+            if indices is not None:
+                indices[i, : len(best_idx)] = torch.tensor(best_idx)
+
             if sent_lengths[i] < sent_max_len:
                 decoded[i, sent_lengths[i]] = eos_token_id
 
         return UserDict(
             {
                 "sequences": decoded,
                 "sequence_scores": best_scores,
+                "beam_indices": indices,
             }
         )
 
@@ -789,6 +820,7 @@ def finalize(
 
         # prepare for adding eos
         sent_lengths_max = sent_lengths.max().item() + 1
+
         sent_max_len = min(sent_lengths_max, max_length) if max_length is not None else sent_lengths_max
         decoded: torch.LongTensor = input_ids.new(batch_size * self.num_beam_hyps_to_keep, sent_max_len)
         # shorter batches are padded if needed
@@ -801,6 +833,7 @@ def finalize(
             decoded[i, : sent_lengths[i]] = hypo
             if sent_lengths[i] < sent_max_len:
                 decoded[i, sent_lengths[i]] = eos_token_id
+
         return UserDict(
             {
                 "sequences": decoded,
@@ -826,15 +859,15 @@ def __len__(self):
         """
         return len(self.beams)
 
-    def add(self, hyp: torch.LongTensor, sum_logprobs: float):
+    def add(self, hyp: torch.LongTensor, sum_logprobs: float, beam_indices: Optional[torch.LongTensor] = None):
         """
         Add a new hypothesis to the list.
         """
         score = sum_logprobs / (hyp.shape[-1] ** self.length_penalty)
         if len(self) < self.num_beams or score > self.worst_score:
-            self.beams.append((score, hyp))
+            self.beams.append((score, hyp, beam_indices))
             if len(self) > self.num_beams:
-                sorted_next_scores = sorted([(s, idx) for idx, (s, _) in enumerate(self.beams)])
+                sorted_next_scores = sorted([(s, idx) for idx, (s, _, _) in enumerate(self.beams)])
                 del self.beams[sorted_next_scores[0][1]]
                 self.worst_score = sorted_next_scores[1][0]
             else: