diff --git a/examples/distillation/training.py b/examples/distillation/training.py
index 34626e7..a42225e 100644
--- a/examples/distillation/training.py
+++ b/examples/distillation/training.py
@@ -95,7 +95,7 @@ def convert_checkpoints_to_hf(model_training_config, output_path, best_model_pat
 def train():
     pl.seed_everything(42)
     parser = HfArgumentParser((TrainingArgs, DataLoadingConfig, DistillTrainingConfig))
-    (model_training_config, data_config, distill_config, _) = parser.parse_args_into_dataclasses(
+    model_training_config, data_config, distill_config, _ = parser.parse_args_into_dataclasses(
         return_remaining_strings=True
     )
     if (
diff --git a/examples/pruning/main_pruning.py b/examples/pruning/main_pruning.py
index 36647fb..84b05a1 100644
--- a/examples/pruning/main_pruning.py
+++ b/examples/pruning/main_pruning.py
@@ -11,7 +11,7 @@
 if __name__ == "__main__":
 
     parser = HfArgumentParser((PruningConfig, CalibrationDataConfig))
-    (pruning_config, data_config) = parser.parse_args_into_dataclasses()
+    pruning_config, data_config = parser.parse_args_into_dataclasses()
     logger.info(f"pruning_config = {pruning_config}")
     logger.info(f"data_config = {data_config}")
 
diff --git a/examples/quantization/main_quantization.py b/examples/quantization/main_quantization.py
index 4ce8ecd..44a130a 100644
--- a/examples/quantization/main_quantization.py
+++ b/examples/quantization/main_quantization.py
@@ -11,7 +11,7 @@
 if __name__ == "__main__":
 
     parser = HfArgumentParser((QuantizationConfig, CalibrationDataConfig))
-    (quantization_config, data_config) = parser.parse_args_into_dataclasses()
+    quantization_config, data_config = parser.parse_args_into_dataclasses()
     logger.info(f"quantization_config = {quantization_config}")
     logger.info(f"data_config = {data_config}")
 
diff --git a/examples/structured_pruning/main_structured_pruning.py b/examples/structured_pruning/main_structured_pruning.py
index e3bfa3c..c6673f4 100644
--- a/examples/structured_pruning/main_structured_pruning.py
+++ b/examples/structured_pruning/main_structured_pruning.py
@@ -11,7 +11,7 @@
 if __name__ == "__main__":
 
     parser = HfArgumentParser((StructuredPruningConfig, CalibrationDataConfig))
-    (pruning_config, data_config) = parser.parse_args_into_dataclasses()
+    pruning_config, data_config = parser.parse_args_into_dataclasses()
     logger.info(f"pruning_config = {pruning_config}")
     logger.info(f"data_config = {data_config}")
 
diff --git a/src/fmchisel/distillation/config.py b/src/fmchisel/distillation/config.py
index dc2bfa3..3aa89a9 100644
--- a/src/fmchisel/distillation/config.py
+++ b/src/fmchisel/distillation/config.py
@@ -73,13 +73,11 @@ class DistillTrainingConfig:
     sample_method: Literal["supervised", "on-policy", "sequence-level"] = field(default="supervised")
     sample_fraction: float = field(
         default=1.0,
-        metadata={
-            "help": "Fraction of batches whose responses are sampled from student (on-policy) distribution \
+        metadata={"help": "Fraction of batches whose responses are sampled from student (on-policy) distribution \
                 or teacher (sequence-evel)  distribution rather than using the original responses, \
                 same as the huggingface GKD trainer (parameter self.lmbda). https://huggingface.co/docs/trl/gkd_trainer#trl.GKDConfig \
                 e.g., 0.4 means 40% of batches are using the responses sampled from student/teacher model, with 60% using original data \
-                Ignored when using supervised methods (ground-truth tokens)."
-        },
+                Ignored when using supervised methods (ground-truth tokens)."},
     )
     max_new_tokens: int = field(
         default=100,
@@ -89,10 +87,8 @@ class DistillTrainingConfig:
     )
     sample_temperature: float = field(
         default=0.8,
-        metadata={
-            "help": "Sample temperature used for on-policy or sequence-level response token generation. \
-                The higher the temperature, the more random the completions."
-        },
+        metadata={"help": "Sample temperature used for on-policy or sequence-level response token generation. \
+                The higher the temperature, the more random the completions."},
     )
     # [end] sampling and generation configs
     include_prompt_loss: bool = field(