fixed annotations

data.py

@@ -1,3 +1,5 @@
+from __future__ import annotations
+
 import dataclasses
 import enum
 import functools
@@ -229,7 +231,7 @@ class AnnotatedImage:
 
 def generate_annotated_images():
     for image_id in tqdm.autonotebook.tqdm(
-        load_train_image_ids(), "Iterating over annotated images"
+            load_train_image_ids(), "Iterating over annotated images"
     ):
         yield AnnotatedImage.from_image_id(image_id)
 
@@ -331,7 +333,7 @@ def convert_number_to_scientific_string(value: int or float) -> str:
 
 
 def convert_axis_data_to_string(
-    axis_data: list[str or float], values_type: ValuesType
+        axis_data: list[str or float], values_type: ValuesType
 ) -> str:
     formatted_axis_data = []
     for value in axis_data:

metrics.py

@@ -1,3 +1,5 @@
+from __future__ import annotations
+
 import numpy as np
 import rapidfuzz
 import sklearn

model.py

@@ -1,3 +1,5 @@
+from __future__ import annotations
+
 import collections
 import dataclasses
 import types
@@ -28,7 +30,7 @@ class Model:
 
 
 def add_unknown_tokens_to_tokenizer(
-    tokenizer, encoder_decoder, unknown_tokens: list[str]
+        tokenizer, encoder_decoder, unknown_tokens: list[str]
 ):
     tokenizer.add_tokens(unknown_tokens)
     encoder_decoder.decoder.resize_token_embeddings(len(tokenizer))
@@ -51,7 +53,7 @@ def find_unknown_tokens_for_tokenizer(tokenizer) -> collections.Counter:
 
 
 def replace_pad_token_id_with_negative_hundred_for_hf_transformers_automatic_batch_transformation(
-    tokenizer, token_ids
+        tokenizer, token_ids
 ):
     token_ids[token_ids == tokenizer.pad_token_id] = -100
     return token_ids
@@ -142,7 +144,7 @@ def build_model(config: types.SimpleNamespace or object) -> Model:
 
 
 def generate_token_strings(
-    model: Model, images: torch.Tensor, skip_special_tokens=True
+        model: Model, images: torch.Tensor, skip_special_tokens=True
 ) -> list[str]:
     decoder_output = model.encoder_decoder.generate(
         images,

train.py

@@ -1,3 +1,5 @@
+from __future__ import annotations
+
 import os
 
 import pandas as pd
@@ -20,12 +22,12 @@ from utils import set_tokenizers_parallelism, set_torch_device_order_pci_bus
 
 class MetricsCallback(pl.callbacks.Callback):
     def on_validation_batch_start(
-        self, trainer, pl_module, batch: Batch, batch_idx, dataloader_idx=0
+            self, trainer, pl_module, batch: Batch, batch_idx, dataloader_idx=0
     ):
         predicted_strings = generate_token_strings(pl_module.model, images=batch.images)
 
         for expected_data_index, predicted_string in zip(
-            batch.data_indices, predicted_strings, strict=True
+                batch.data_indices, predicted_strings, strict=True
         ):
             benetech_score = benetech_score_string_prediction(
                 expected_data_index=expected_data_index,
@@ -50,7 +52,7 @@ class MetricsCallback(pl.callbacks.Callback):
 
 class TransformersPreTrainedModelsCheckpointIO(pl.plugins.CheckpointIO):
     def __init__(
-        self, pretrained_models: list[transformers.modeling_utils.PreTrainedModel]
+            self, pretrained_models: list[transformers.modeling_utils.PreTrainedModel]
     ):
         super().__init__()
         self.pretrained_models = pretrained_models

utils.py

@@ -1,3 +1,5 @@
+from __future__ import annotations
+
 import os
 import pickle
 from typing import Callable, TypeVar
@@ -14,7 +16,7 @@ def set_torch_device_order_pci_bus():
 
 
 def load_pickle_or_build_object_and_save(
-    pickle_path: str, build_object: Callable[[], T], overwrite=False
+        pickle_path: str, build_object: Callable[[], T], overwrite=False
 ) -> T:
     if overwrite or not os.path.exists(pickle_path):
         pickle.dump(build_object(), open(pickle_path, "wb"))