reformat

keras_cv_attention_models/backend.py

@@ -16,14 +16,14 @@
  try:
  from importlib.metadata import version
 
- tf_version = version('tensorflow').split(".")
+ tf_version = version("tensorflow").split(".")
  try:
- keras_version = version('keras').split(".")
+ keras_version = version("keras").split(".")
  except:
  keras_version = tf_version
 
  try:
- tf_keras_version = version('tf_keras').split(".")
+ tf_keras_version = version("tf_keras").split(".")
  except:
  tf_keras_version = keras_version
 

keras_cv_attention_models/clip/tokenizer.py

@@ -3,6 +3,7 @@
 Copied from https://github.com/mlfoundations/open_clip/blob/main/src/open_clip/tokenizer.py
 Copied from https://github.com/openai/CLIP. Originally MIT License, Copyright (c) 2021 OpenAI.
 """
+
 import os
 import html
 from functools import lru_cache

keras_cv_attention_models/coco/eval_func.py

@@ -177,7 +177,7 @@ def __decode_single__(
  self.__init_anchor__(input_shape)
 
  if self.num_masks > 0: # Segmentation masks
- pred, masks = pred[:, :-self.num_masks], pred[:, -self.num_masks:]
+ pred, masks = pred[:, : -self.num_masks], pred[:, -self.num_masks :]
  else:
  masks = None
 

keras_cv_attention_models/coco/torch_losses.py

@@ -1,6 +1,7 @@
 """
 Just copied from ultralytics tal.py / Loss function / BboxLoss
 """
+
 import torch
 import math
 from torch import nn
@@ -30,9 +31,7 @@ def __init__(self, device="cpu", stride=[8, 16, 32], nc=80, reg_max=16, input_sh
  self.proj = torch.arange(reg_max, dtype=torch.float, device=self.device)
 
  def get_feature_sizes(self):
- feature_sizes = [
- (math.ceil(self.input_shape[-2] / (2**ii)), math.ceil(self.input_shape[-1] / (2**ii))) for ii in range(3, 3 + self.num_pyramid_levels)
- ]
+ feature_sizes = [(math.ceil(self.input_shape[-2] / (2**ii)), math.ceil(self.input_shape[-1] / (2**ii))) for ii in range(3, 3 + self.num_pyramid_levels)]
  feature_lens = [ii[0] * ii[1] for ii in feature_sizes]
  return feature_sizes, feature_lens
 

keras_cv_attention_models/efficientnet/efficientnet_edgetpu.py

@@ -1,6 +1,7 @@
 """ Creates an EfficientNet-EdgeTPU model
 Ref impl: https://github.com/tensorflow/tpu/tree/master/models/official/efficientnet/edgetpu
 """
+
 import math
 from keras_cv_attention_models.efficientnet.efficientnet_v2 import EfficientNetV2
 from keras_cv_attention_models.attention_layers import make_divisible

keras_cv_attention_models/efficientnet/efficientnet_v1.py

@@ -2,6 +2,7 @@
 Creates a EfficientNetV1 Model as defined in: EfficientNetV1: Self-training with Noisy Student improves ImageNet classification.
 arXiv preprint arXiv:1911.04252.
 """
+
 import math
 from keras_cv_attention_models.efficientnet.efficientnet_v2 import EfficientNetV2
 from keras_cv_attention_models.models import register_model

keras_cv_attention_models/efficientnet/efficientnet_v2.py

@@ -2,6 +2,7 @@
 Creates a EfficientNetV2 Model as defined in: Mingxing Tan, Quoc V. Le. (2021). arXiv preprint arXiv:2104.00298.
 EfficientNetV2: Smaller Models and Faster Training.
 """
+
 import numpy as np
 from keras_cv_attention_models import backend
 from keras_cv_attention_models.backend import layers, functional, models, is_channels_last

keras_cv_attention_models/gpt2/gpt2.py

@@ -61,7 +61,7 @@ def call(self, inputs):
  else:
  causal_mask = self.causal_mask
  qq_len, kk_len = (functional.shape(inputs)[2], functional.shape(inputs)[3]) if backend.is_tensorflow_backend else (inputs.shape[2], inputs.shape[3])
- return (inputs + causal_mask[:, :, :qq_len, :kk_len])
+ return inputs + causal_mask[:, :, :qq_len, :kk_len]
 
  def compute_output_shape(self, input_shape):
  return input_shape[0] if self.is_kv_cache else input_shape

keras_cv_attention_models/imagenet/augment.py

@@ -2,6 +2,7 @@
 Copied from: https://github.com/tensorflow/models/blob/master/official/vision/image_classification/augment.py
 Midified according to: https://github.com/rwightman/pytorch-image-models/blob/master/timm/data/auto_augment.py
 """
+
 # Copyright 2019 The TensorFlow Authors. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -724,7 +725,7 @@ def select_and_apply_random_policy(policies: Any, image: tf.Tensor):
  policy_to_select = tf.random.uniform([], maxval=len(policies), dtype=tf.int32)
  # Note that using tf.case instead of tf.conds would result in significantly
  # larger graphs and would even break export for some larger policies.
- for (i, policy) in enumerate(policies):
+ for i, policy in enumerate(policies):
  image = tf.cond(tf.equal(i, policy_to_select), lambda selected_policy=policy: selected_policy(image), lambda: image)
  return image
 
@@ -927,6 +928,7 @@ def __call__(self, image: tf.Tensor) -> tf.Tensor:
  for policy_info in policy:
  policy_info = list(policy_info) + [replace_value, self.cutout_const, self.translate_const]
  tf_policy.append(_parse_policy_info(*policy_info))
+
  # Now build the tf policy that will apply the augmentation procedue
  # on image.
  def make_final_policy(tf_policy_):

keras_cv_attention_models/imagenet/callbacks.py

@@ -384,7 +384,7 @@ def set_model(self, model):
  param.requires_grad_(False)
  self.basic_save_name = model.name if self.basic_save_name is None else self.basic_save_name
  self.save_file_path = os.path.join(self.save_path, self.basic_save_name) + "_ema.h5"
- self.ema.history = namedtuple('history', ['history'])({}) # Record history and pass back to model if any
+ self.ema.history = namedtuple("history", ["history"])({}) # Record history and pass back to model if any
  self.enabled = True
 
  def on_train_batch_end(self, batch, logs=None):

keras_cv_attention_models/imagenet/train_func.py

@@ -31,8 +31,8 @@ def set_random_seed(seed):
  tf.random.set_seed(seed)
  tf.experimental.numpy.random.seed(seed)
  # When running on the CuDNN backend, two further options must be set
- os.environ['TF_CUDNN_DETERMINISTIC'] = '1'
- os.environ['TF_DETERMINISTIC_OPS'] = '1'
+ os.environ["TF_CUDNN_DETERMINISTIC"] = "1"
+ os.environ["TF_DETERMINISTIC_OPS"] = "1"
 
 
 def init_global_strategy(enable_float16=True, seed=0, TPU=False):

keras_cv_attention_models/llama2/llama2.py

@@ -48,7 +48,7 @@ def call(self, inputs, **kwargs):
  cur_tensor, start_pos = inputs, 0
  left, right = functional.unstack(cur_tensor, axis=-2)
  seq_len = functional.shape(left)[-3] if backend.is_tensorflow_backend else left.shape[-3]
- pos_cos, pos_sin = self.pos_cos[start_pos:start_pos + seq_len], self.pos_sin[start_pos:start_pos + seq_len]
+ pos_cos, pos_sin = self.pos_cos[start_pos : start_pos + seq_len], self.pos_sin[start_pos : start_pos + seq_len]
  out = functional.stack([left * pos_cos - right * pos_sin, right * pos_cos + left * pos_sin], axis=-2)
  return out
 
@@ -93,6 +93,7 @@ class KVCache(layers.Layer):
  print(f"{np.allclose(aa.cache[:1, :4], out) = }")
  # np.allclose(aa.cache[:1, :4], out) = True
  """
+
  def __init__(self, max_batch_size=32, max_seq_len=2048, **kwargs):
  super().__init__(**kwargs)
  self.max_batch_size, self.max_seq_len = max_batch_size, max_seq_len
@@ -113,7 +114,7 @@ def call(self, inputs):
  else:
  batch_size, seq_len = functional.shape(cur_tensor)[0], functional.shape(cur_tensor)[1]
  tensor_with_cache = functional.concat([self.cache[:batch_size, :start_pos], cur_tensor], axis=1)
- cur_batch_cache = functional.concat([tensor_with_cache, self.cache[:batch_size, start_pos + seq_len:]], axis=1)
+ cur_batch_cache = functional.concat([tensor_with_cache, self.cache[:batch_size, start_pos + seq_len :]], axis=1)
  self.cache.assign(functional.concat([cur_batch_cache, self.cache[batch_size:]], axis=0))
  return tensor_with_cache
 

keras_cv_attention_models/pytorch_backend/initializers.py

@@ -8,7 +8,7 @@ def _to_dtype_(tensor, dtype=None):
  if dtype is None:
  return tensor
 
- default_dtype = str(torch.get_default_dtype()).split('.')[-1] # torch.float32 -> "float32"
+ default_dtype = str(torch.get_default_dtype()).split(".")[-1] # torch.float32 -> "float32"
  if dtype == default_dtype:
  return tensor
  return tensor.to(dtype=getattr(torch, dtype, torch.get_default_dtype()))

keras_cv_attention_models/pytorch_backend/models.py

@@ -69,7 +69,7 @@ def train_step(self, xx, yy):
  def fit(self, x=None, y=None, batch_size=32, epochs=1, callbacks=None, validation_data=None, initial_epoch=0, validation_batch_size=None, **kwargs):
  callbacks = callbacks or []
  [ii.set_model(self) for ii in callbacks if ii.model is None]
- self.history = namedtuple('history', ['history', 'epoch'])({"loss": []}, []) # Mimic of `keras.callbacks.History`
+ self.history = namedtuple("history", ["history", "epoch"])({"loss": []}, []) # Mimic of `keras.callbacks.History`
  validation_batch_size = validation_batch_size or batch_size
  self.batch_size, self.callbacks, self.validation_batch_size = batch_size, callbacks, validation_batch_size
  self.accumulate_passed_batches = 0

keras_cv_attention_models/volo/volo.py

@@ -303,6 +303,7 @@ def __init__(self, scale=2, beta=1.0, **kwargs):
 
  def call(self, inputs, training=None, **kwargs):
  height, width = functional.shape(inputs)[1], functional.shape(inputs)[2]
+
  # tf.print("training:", training)
  def _call_train():
  return functional.stack(self.rand_bbox(height, width))

keras_cv_attention_models/yolov8/train.py

@@ -1,5 +1,6 @@
 import numpy as np
 
+
 class ModelEMA:
  """Updated Exponential Moving Average (EMA) from https://github.com/rwightman/pytorch-image-models
  Keeps a moving average of everything in the model state_dict (parameters and buffers)