[Fix] utf-8 codec cant decode

Author: huyiwen

utilization/dataset/dataset.py

@@ -766,15 +766,16 @@ def log_final_results(
         return log_final_results(
             raw_predictions=raw_predictions,
             processed_predictions=processed_predictions,
+            evaluation_instances=self.evaluation_instances,
             score_lists=score_lists,
             multiple_source=(self.dataset_name == "winogrande"),
             model_evaluation_method=self.model_evaluation_method,
             use_normalization=self.use_normalization,
             option_nums=self.option_nums,
             len_evaluation_data=len(self.evaluation_data),
-            evaluation_instances=self.evaluation_instances,
             sample_num=self.sample_num,
             references=self.references,
+            local_model=self.model.is_local_model(),
         )
 
     def __repr__(self):
@@ -967,7 +968,7 @@ def step(
             if batch_size > 0:
                 tqdm.set_description(self.display_names[self._cur_idx])
         if batch_size > 0:
-            writer.log_batch_results(batch_raw_predictions, self._lines_iter)
+            writer.log_batch_results(batch_raw_predictions, self._datasets[0].model.is_local_model(), self._lines_iter)
 
     def __repr__(self):
         reprs = []

utilization/model/model.py

@@ -248,8 +248,10 @@ def set_generation_args(self, **extra_model_args):
             logger.warning(f"Unused generation arguments: {extra_model_args}")
         return self.generation_kwargs
 
-    def generation(self, batched_inputs: Union[List[str],
-                                               List[Conversation]]) -> Union[List[str], List[Tuple[str, ...]]]:
+    def generation(
+        self,
+        batched_inputs: Union[List[str], List[Conversation]],
+    ) -> Union[List[str], List[Tuple[str, ...]]]:
         multi_turn_results = self.request(
             prompt=batched_inputs,
             multi_turn=self.multi_turn,

utilization/utils/arguments.py

@@ -690,14 +690,15 @@ def parse_argument(args: Optional[List[str]] = None,
         epilog=EXAMPLE_STRING,
         formatter_class=argparse.RawDescriptionHelpFormatter,
     )
-    model_args, dataset_args, evaluation_args = parser.parse_args_into_dataclasses(args)
 
     try:
         from dotenv import load_dotenv
         load_dotenv()
     except (ImportError, ModuleNotFoundError):
         pass
 
+    model_args, dataset_args, evaluation_args = parser.parse_args_into_dataclasses(args)
+
     if model_args.bnb_config:
         bnb_config_dict = json.loads(model_args.bnb_config)
         model_args.bnb_config = BitsAndBytesConfig(**bnb_config_dict)

utilization/utils/conversation.py

@@ -331,6 +331,9 @@ def to_model_prompt(
             max_turns=max_turns,
         )[0]
 
+    def apply_prompt_template(self):
+        return self.formatter.apply_prompt_template(self)
+
     def add(
         self,
         other: Optional["Conversation"] = None,

utilization/utils/log_results.py

@@ -6,6 +6,8 @@
 
 import pandas as pd
 
+from .conversation import Conversation
+
 logger = getLogger(__name__)
 
 if typing.TYPE_CHECKING:
@@ -35,6 +37,22 @@ def wrapper(df: pd.DataFrame):
     return wrapper
 
 
+def dump_conversations(convs: List[Any], local: bool):
+    if isinstance(convs, (str, Conversation)):
+        convs = [convs]
+
+    if isinstance(convs[0], Conversation):
+        if not local:
+            convs = [[str(m['content']) for m in p.messages] for p in convs]
+        else:
+            convs = [p.apply_prompt_template() for p in convs]
+
+    if not isinstance(convs[0], str):
+        convs = [str(p) for p in convs]
+
+    return convs
+
+
 class PredictionWriter:
 
     def __init__(self, evaluation_path: Optional[str]):
@@ -83,14 +101,15 @@ def _write(self, data):
     def log_batch_results(
         self,
         raw_predictions: List[str],
+        local_model: bool,
         lines_iter: Iterator[Tuple[int, str, Any]],
     ) -> int:
+        """Log the batch predictions to the evaluation jsonlines file."""
         if not self.alive():
             return len(raw_predictions)
 
         for raw_prediction, (idx, source, reference) in zip(raw_predictions, lines_iter):
-            if not isinstance(source, str):
-                source = str(source)
+            source = dump_conversations(source, local_model)
             lines = {
                 "index": idx,
                 "source": source,
@@ -140,16 +159,20 @@ def load_continue(self) -> Iterator[typing.Any]:
 def log_final_results(
     raw_predictions: List[str],
     processed_predictions: List[Union[str, float]],
+    evaluation_instances: List[tuple],
     score_lists: Dict[str, List[float]],
     multiple_source: bool,
     model_evaluation_method: str,
     use_normalization: bool,
     option_nums: List[int],
     len_evaluation_data: int,
-    evaluation_instances: List[tuple],
     sample_num: int,
     references: List[Any],
+    local_model: bool,
 ) -> Optional[pd.Series]:
+    """Aggregate the final results and prepare for dumping to a json file."""
+
+    evaluation_instances = dump_conversations(evaluation_instances, local_model)
 
     transposed_score_lists = [dict(zip(score_lists.keys(), values)) for values in zip(*score_lists.values())]
     if model_evaluation_method == "generation":