From 0f9365a14f290066d1ee1be30a4b6e2875279bbc Mon Sep 17 00:00:00 2001
From: philheller <philip.heller@student.kit.edu>
Date: Fri, 6 Sep 2024 12:45:22 +0200
Subject: [PATCH] Cleanup, comment out some print statements

---
 src/transformers/generation/utils.py | 29 ++++++++++------------------
 1 file changed, 10 insertions(+), 19 deletions(-)

diff --git a/src/transformers/generation/utils.py b/src/transformers/generation/utils.py
index 19535b23d950..723a3feba754 100644
--- a/src/transformers/generation/utils.py
+++ b/src/transformers/generation/utils.py
@@ -276,6 +276,7 @@ class GenerateBeamDecoderOnlyOutput(ModelOutput):
     # for continuation
     attention_mask: Optional[torch.LongTensor] = None
     last_beam_scores: Optional[torch.FloatTensor] = None
+    # group beam search
     next_input_ids: Optional[torch.LongTensor] = None
 
 
@@ -341,6 +342,7 @@ class GenerateBeamEncoderDecoderOutput(ModelOutput):
     # for continuation
     attention_mask: Optional[torch.LongTensor] = None
     last_beam_scores: Optional[torch.FloatTensor] = None
+    # group beam search
     next_input_ids: Optional[torch.LongTensor] = None
 
 
@@ -1727,9 +1729,9 @@ def generate(
         tokenizer = kwargs.pop("tokenizer", None)  # Pull this out first, we only use it for stopping criteria
         generation_config, model_kwargs = self._prepare_generation_config(generation_config, **kwargs)
         # todo:phil remove in prod
-        print(generation_config, 
-                {k: v for k, v in model_kwargs.items() if k != 'past_key_values'}
-                ) 
+        # print(generation_config, 
+        #         {k: v for k, v in model_kwargs.items() if k != 'past_key_values'}
+        #         ) 
         self._validate_model_kwargs(model_kwargs.copy())
         self._validate_assistant(assistant_model)
 
@@ -1935,7 +1937,7 @@ def generate(
         # todo:phil figure out for every mode, what continuation would look like
         # ? input for every configuration step would have to be the output from the previous step
         # ! leaving out speculative decoding for now
-        print("GENERATION_MODE", generation_mode)
+        # print("GENERATION_MODE", generation_mode)
 
         if generation_mode == GenerationMode.ASSISTED_GENERATION:
             if generation_config.num_return_sequences > 1:
@@ -2190,9 +2192,9 @@ def generate(
                 )
 
         elif generation_mode == GenerationMode.CONSTRAINED_BEAM_SEARCH:
-            if generation_config.resume_generation:
+            # if generation_config.resume_generation:
                 # todo:phil implement
-                print("Resuming generation")
+                # print("Resuming generation")
             final_constraints = []
             if generation_config.constraints is not None:
                 final_constraints = generation_config.constraints
@@ -2993,12 +2995,8 @@ def _sample(
         generation_config: GenerationConfig,
         synced_gpus: bool,
         streamer: Optional["BaseStreamer"],
-<<<<<<< HEAD
-        logits_warper: Optional[LogitsProcessorList] = None,
-        last_scores: Optional[Tuple[torch.FloatTensor]] = None,
-=======
         logits_warper: Optional[LogitsProcessorList],
->>>>>>> 7789317cb66e3040bb99a9359259d750e30a276f
+        last_scores: Optional[Tuple[torch.FloatTensor]] = None,
         **model_kwargs,
     ) -> Union[GenerateNonBeamOutput, torch.LongTensor]:
         r"""
@@ -3204,7 +3202,6 @@ def _temporary_reorder_cache(self, past_key_values, beam_idx):
             past_key_values.reorder_cache(beam_idx)
         return past_key_values
 
-<<<<<<< HEAD
     def _reorder_attention_mask(self, beam_indices: Tuple[Tuple[int]], **model_kwargs):
         last_tokens = [inner_tuple[-1] for inner_tuple in beam_indices]
         last_indices = torch.tensor([token.item() for token in last_tokens])
@@ -3219,8 +3216,6 @@ def _reorder_attention_mask(self, beam_indices: Tuple[Tuple[int]], **model_kwarg
         return model_kwargs
 
     # TODO (joao, v4.42): remove default for `logits_warper`
-=======
->>>>>>> 7789317cb66e3040bb99a9359259d750e30a276f
     def _beam_search(
         self,
         input_ids: torch.LongTensor,
@@ -3229,14 +3224,10 @@ def _beam_search(
         stopping_criteria: StoppingCriteriaList,
         generation_config: GenerationConfig,
         synced_gpus: bool,
-<<<<<<< HEAD
-        logits_warper: Optional[LogitsProcessorList] = None,
+        logits_warper: Optional[LogitsProcessorList],
         last_beam_scores: Optional[torch.FloatTensor] = None,
         last_scores: Optional[Tuple[torch.FloatTensor]] = None,
         original_prompt_len: int = None,
-=======
-        logits_warper: Optional[LogitsProcessorList],
->>>>>>> 7789317cb66e3040bb99a9359259d750e30a276f
         **model_kwargs,
     ) -> Union[GenerateBeamOutput, torch.LongTensor]:
         r"""