style: format

.flake8

@@ -49,6 +49,10 @@ extend-ignore =
 		# Google Python style is not RST until after processed by Napoleon
 		# See https://github.com/peterjc/flake8-rst-docstrings/issues/17
 		RST201,RST203,RST301,
+		# It happens too often
+		C416, C419,
+		# This is new and cba to change the repo
+		S113
 extend-select =
 	# Should raise AssertionError instead of assert False
 	B011,
@@ -69,8 +73,6 @@ extend-select =
 	# Within an except clause, raise exceptions with `raise ... from err` or `raise ...
 	# from None` to distinguish them from errors in exception handling
 	B904,
-	# Alternative to E501 regarding line length
-	B950,
 	#	Counterpart to W503, enforce having the operator at the start of a new line.
 	W504,
 

src/emma_policy/api/clients/feature_extractor.py

@@ -70,7 +70,7 @@ def extract_single_image(self, image: Union[Image.Image, ArrayLike]) -> FeatureR
         try:
             response.raise_for_status()
         except requests.exceptions.HTTPError as err:
-            raise SystemExit(err)
+            raise SystemExit(err) from err
 
         data = response.json()
         feature_response = FeatureResponse(
@@ -103,7 +103,7 @@ def extract_batch_images(
         try:
             response.raise_for_status()
         except requests.exceptions.HTTPError as err:
-            raise SystemExit(err)
+            raise SystemExit(err) from err
 
         data = response.json()
 

src/emma_policy/datamodules/coco_captioning_dataset.py

@@ -29,7 +29,6 @@ def __init__(
         merged_annotations: bool = True,
         is_train: bool = True,
     ) -> None:
-
         if not merged_annotations:
             raise NotImplementedError(
                 "Expecting dbs where every instance is an image associated with all of its captions."

src/emma_policy/datamodules/nlvr2_dataset.py

@@ -27,7 +27,6 @@ def __init__(
         max_frames: int = 0,
         use_task_prefix: bool = False,
     ) -> None:
-
         super().__init__(
             dataset_db_path=dataset_db_path, tokenizer=tokenizer, max_frames=max_frames
         )

src/emma_policy/datamodules/pretrain_dataset.py

@@ -833,7 +833,6 @@ def _region_mapping(
         width: int,
         height: int,
     ) -> tuple[torch.Tensor, torch.Tensor]:
-
         gt_bbox = []
         for region in regions:
             gt_bbox_coord = BoxMode.convert(
@@ -894,7 +893,6 @@ def _convert_trajectory_to_text(
             trajectory_text.extend(split_action_name(action.api_action.action))
             # Match the object to a predicted bounding box
             if "bbox" in action.discrete_action.args:
-
                 bbox_coord = action.discrete_action.args["bbox"]  # noqa: WPS529
                 gt_bbox = torch.tensor(
                     [

src/emma_policy/datamodules/refcoco_dataset.py

@@ -35,7 +35,6 @@ def __init__(
         shuffle_objects: bool = False,
         train_with_golden_bbox_prob: float = 1.0,
     ) -> None:
-
         super().__init__(
             dataset_db_path=dataset_db_path,
             tokenizer=tokenizer,

src/emma_policy/datamodules/vqa_v2_dataset.py

@@ -27,7 +27,6 @@ def __init__(
         tokenizer: PreTrainedTokenizer,
         max_frames: int = 0,
     ) -> None:
-
         super().__init__(
             dataset_db_path=dataset_db_path,
             tokenizer=tokenizer,

src/emma_policy/models/decoder_emma.py

@@ -50,7 +50,6 @@ def decoder_layer_outputs(
     ) -> tuple[torch.FloatTensor, ...]:
         """Get output from a single decoder layer."""
         if self.gradient_checkpointing and self.training:
-
             if use_cache:
                 logger.warning(
                     "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."

src/emma_policy/models/loss_utils.py

@@ -9,7 +9,8 @@ def tiny_value_of_dtype(dtype: torch.dtype) -> float:
 
     This is used to avoid numerical issues such as division by zero. This is different from
     `info_value_of_dtype(dtype).tiny` because it causes some NaN bugs. Only supports floating point
-    dtypes. Implementation from AllenNLP: https://github.com/allenai/allennlp/blob/39c40fe38cd2fd36b3465b0b3c031f54ec824160/allennlp/nn/util.py#L2010-L2024
+    dtypes. Implementation from AllenNLP:
+    https://github.com/allenai/allennlp/blob/39c40fe38cd2fd36b3465b0b3c031f54ec824160/allennlp/nn/util.py#L2010-L2024
     """
     if not dtype.is_floating_point:
         raise TypeError("Only supports floating point dtypes.")

src/emma_policy/models/model_output_emma.py

@@ -7,7 +7,9 @@
 
 @dataclass
 class EmmaSeq2SeqModelOutput(ModelOutput):
-    """Base class for model encoder's outputs that also contains : pre-computed hidden states that can speed up sequential decoding.
+    """Base class for encoder outputs.
+
+    Also contains pre-computed hidden states that can speed up sequential decoding.
 
     Args:
         last_hidden_state (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`):

src/emma_policy/models/nlvr2_emma_policy.py

@@ -35,7 +35,6 @@ def __init__(
         nlvr2_metrics: Optional[list[str]] = None,
         **kwargs: Any,
     ) -> None:
-
         self._tokenizer = AutoTokenizer.from_pretrained(model_name)
         self._pred_gt: dict[str, list[str]] = {
             "predictions": [],

src/emma_policy/utils/boxes.py

@@ -100,10 +100,10 @@ def _convert(self, box: torch.Tensor, from_mode: BoxMode, to_mode: BoxMode) -> t
 
         try:
             converted_box = convert_functions[from_mode][to_mode](box)
-        except KeyError:
+        except KeyError as err:
             raise NotImplementedError(
                 f"Conversion from BoxMode {from_mode} to {to_mode} is not supported."
-            )
+            ) from err
 
         return converted_box
 

src/emma_policy/utils/masks.py

@@ -13,7 +13,8 @@ def decompress_simbot_mask(
 ) -> Union[torch.Tensor, typing.NDArray[np.float64]]:
     """Decompress a compressed mask array.
 
-    Adopted from https://us-east-1.console.aws.amazon.com/codesuite/codecommit/repositories/AlexaSimbotMLToolbox/browse/refs/heads/main/--/AlexaSimbotToolbox/arena_wrapper/util/__init__.py?region=us-east-1
+    Adopted from
+    https://us-east-1.console.aws.amazon.com/codesuite/codecommit/repositories/AlexaSimbotMLToolbox/browse/refs/heads/main/--/AlexaSimbotToolbox/arena_wrapper/util/__init__.py?region=us-east-1
     """
     mask = np.zeros((image_width, image_height))
     for start_idx, run_len in compressed_mask:
@@ -29,7 +30,8 @@ def compress_simbot_mask(
 ) -> list[list[int]]:
     """Compress a binary 2D array mask for the simbot arena.
 
-    Adopted from https://us-east-1.console.aws.amazon.com/codesuite/codecommit/repositories/AlexaSimbotMLToolbox/browse/refs/heads/main/--/AlexaSimbotToolbox/arena_wrapper/util/__init__.py?region=us-east-1
+    Adopted from
+    https://us-east-1.console.aws.amazon.com/codesuite/codecommit/repositories/AlexaSimbotMLToolbox/browse/refs/heads/main/--/AlexaSimbotToolbox/arena_wrapper/util/__init__.py?region=us-east-1
     """
     # list of lists of run lengths for 1s, which are assumed to be less frequent.
     run_len_compressed: list[list[int]] = []

src/emma_policy/utils/samplers/distributed_weighted_sampler.py

@@ -17,7 +17,6 @@ def __init__(
         rank: Optional[int] = None,
         replacement: bool = True,
     ) -> None:
-
         if num_replicas is None:
             if not torch.distributed.is_available():
                 raise RuntimeError("Requires distributed package to be available")

tests/datamodules/test_datamodule_helpers.py

@@ -48,7 +48,6 @@ def test_simbot_target_tokens(
     action_text: str,
     emma_tokenizer: EmmaTokenizer,
 ) -> None:
-
     target_encoding = emma_tokenizer.encode_plus(target_text, return_tensors="pt", truncation=True)
     full_target_token_ids = target_encoding.input_ids.squeeze(0)
     target_token_ids = mask_past_target_actions(