Deprecate many datasets and models

composer/callbacks/mlperf.py

@@ -267,6 +267,8 @@ def _get_dataloader_stats(self, dataloader: Iterable):
             # attempt to import ffcv and test if its an ffcv loader.
             import ffcv  # type: ignore
 
+            warnings.warn(DeprecationWarning('ffcv is deprecated and will be removed in v0.18'))
+
             if isinstance(dataloader, ffcv.loader.Loader):
                 # Use the cached attribute ffcv.init_traversal_order to compute number of samples
                 return (

composer/datasets/ade20k.py

@@ -8,6 +8,7 @@
 """
 
 import os
+import warnings
 from math import ceil
 from typing import Any, Dict, Optional, Tuple, Union
 
@@ -50,6 +51,8 @@ def build_ade20k_transformations(split,
            image_transforms (torch.nn.Module): Transformations to apply to the input image only.
            target_transforms (torch.nn.Module): Transformations to apply to the target semantic segmentation mask only.
     """
+    warnings.warn(DeprecationWarning('build_ade20k_transformations is deprecated and will be removed in v0.18'))
+
     if split == 'train':
         both_transforms = torch.nn.Sequential(
             RandomResizePair(
@@ -110,6 +113,8 @@ def build_ade20k_dataloader(
             Default: ``true``.
         **dataloader_kwargs (Dict[str, Any]): Additional settings for the dataloader (e.g. num_workers, etc.)
     """
+    warnings.warn(DeprecationWarning('build_ade20k_dataloader is deprecated and will be removed in v0.18'))
+
     if global_batch_size % dist.get_world_size() != 0:
         raise ValueError(
             f'global_batch_size ({global_batch_size}) must be divisible by world_size ({dist.get_world_size()}).')
@@ -195,6 +200,8 @@ def build_streaming_ade20k_dataloader(
             Defaults to ``None``, which is interpreted as the number of nodes of the initial run.
         **dataloader_kwargs (Dict[str, Any]): Additional settings for the dataloader (e.g. num_workers, etc.)
     """
+    warnings.warn(DeprecationWarning('build_streaming_ade20k_dataloader is deprecated and will be removed in v0.18'))
+
     if global_batch_size % dist.get_world_size() != 0:
         raise ValueError(
             f'global_batch_size ({global_batch_size}) must be divisible by world_size ({dist.get_world_size()}).')
@@ -274,6 +281,8 @@ def build_synthetic_ade20k_dataloader(
         memory_format (:class:`composer.core.MemoryFormat`): Memory format of the tensors. Default: ``CONTIGUOUS_FORMAT``.
         **dataloader_kwargs (Dict[str, Any]): Additional settings for the dataloader (e.g. num_workers, etc.)
     """
+    warnings.warn(DeprecationWarning('build_synthetic_ade20k_dataloader is deprecated and will be removed in v0.18'))
+
     if global_batch_size % dist.get_world_size() != 0:
         raise ValueError(
             f'global_batch_size ({global_batch_size}) must be divisible by world_size ({dist.get_world_size()}).')
@@ -321,6 +330,7 @@ def __init__(self, min_scale: float, max_scale: float, base_size: Optional[Tuple
         self.min_scale = min_scale
         self.max_scale = max_scale
         self.base_size = base_size
+        warnings.warn(DeprecationWarning('RandomResizePair is deprecated and will be removed in v0.18'))
 
     def forward(self, sample: Tuple[Image.Image, Image.Image]):
         image, target = sample
@@ -348,6 +358,7 @@ def __init__(self, crop_size: Tuple[int, int], class_max_percent: float = 1.0, n
         self.crop_size = crop_size
         self.class_max_percent = class_max_percent
         self.num_retry = num_retry
+        warnings.warn(DeprecationWarning('RandomCropPair is deprecated and will be removed in v0.18'))
 
     def forward(self, sample: Tuple[Image.Image, Image.Image]):
         image, target = sample
@@ -392,6 +403,7 @@ class RandomHFlipPair(torch.nn.Module):
     def __init__(self, probability: float = 0.5):
         super().__init__()
         self.probability = probability
+        warnings.warn(DeprecationWarning('RandomHFlipPair is deprecated and will be removed in v0.18'))
 
     def forward(self, sample: Tuple[Image.Image, Image.Image]):
         image, target = sample
@@ -413,6 +425,7 @@ def __init__(self, size: Tuple[int, int], fill: Union[int, Tuple[int, int, int]]
         super().__init__()
         self.size = size
         self.fill = fill
+        warnings.warn(DeprecationWarning('PadToSize is deprecated and will be removed in v0.18'))
 
     def forward(self, image: Image.Image):
         padding = max(self.size[0] - image.height, 0), max(self.size[1] - image.width, 0)
@@ -440,6 +453,7 @@ def __init__(self, brightness: float, contrast: float, saturation: float, hue: f
         self.contrast = contrast
         self.saturation = saturation
         self.hue = hue
+        warnings.warn(DeprecationWarning('PhotometricDistoration is deprecated and will be removed in v0.18'))
 
     def forward(self, image: Image.Image):
         if np.random.randint(2):
@@ -490,6 +504,7 @@ def __init__(self,
                  both_transforms: Optional[torch.nn.Module] = None,
                  image_transforms: Optional[torch.nn.Module] = None,
                  target_transforms: Optional[torch.nn.Module] = None):
+        warnings.warn(DeprecationWarning('ADE20k is deprecated and will be removed in v0.18'))
         super().__init__()
         self.datadir = datadir
         self.split = split

composer/datasets/brats.py

@@ -10,6 +10,7 @@
 import glob
 import os
 import random
+import warnings
 
 import numpy as np
 import torch
@@ -36,6 +37,8 @@ def build_brats_dataloader(datadir: str,
         global_batch_size (int): Global batch size.
         **dataloader_kwargs (Dict[str, Any]): Additional settings for the dataloader (e.g. num_workers, etc.)
     """
+    warnings.warn(DeprecationWarning('build_brats_dataloader is deprecated and will be removed in v0.18'))
+
     if global_batch_size % dist.get_world_size() != 0:
         raise ValueError(
             f'global_batch_size ({global_batch_size}) must be divisible by world_size ({dist.get_world_size()}).')

composer/datasets/c4.py

@@ -7,6 +7,7 @@
 <https://commoncrawl.org>`_ dataset.
 """
 import logging
+import warnings
 from typing import Any, Dict, Optional
 
 from torch.utils.data import DataLoader
@@ -74,6 +75,7 @@ def build_streaming_c4_dataloader(
             Defaults to ``None``, which is interpreted as the number of nodes of the initial run.
         **dataloader_kwargs (Dict[str, Any]): Additional settings for the dataloader (e.g. num_workers, etc.)
     """
+    warnings.warn(DeprecationWarning('build_streaming_c4_dataloader is deprecated and will be removed in v0.18'))
 
     try:
         import transformers

composer/datasets/cifar.py

@@ -9,6 +9,7 @@
 
 import os
 import textwrap
+import warnings
 from typing import Any, Dict, List, Optional, Union
 
 import torch
@@ -52,6 +53,8 @@ def build_cifar10_dataloader(
         shuffle (bool): Shuffle the dataset. Default: ``True``.
         **dataloader_kwargs (Any): Additional settings for the dataloader (e.g. num_workers, etc.)
     """
+    warnings.warn(DeprecationWarning('build_cifar10_dataloader is deprecated and will be removed in v0.18'))
+
     if global_batch_size % dist.get_world_size() != 0:
         raise ValueError(
             f'global_batch_size ({global_batch_size}) must be divisible by world_size ({dist.get_world_size()}).')
@@ -119,6 +122,8 @@ def build_ffcv_cifar10_dataloader(
         ``False``.
         datadir (str | None, optional): Path to the non-FFCV data directory.
     """
+    warnings.warn(DeprecationWarning('build_ffcv_cifar10_dataloader is deprecated and will be removed in v0.18'))
+
     try:
         import ffcv
         from ffcv.fields.decoders import IntDecoder, SimpleRGBImageDecoder

composer/datasets/ffcv_utils.py

@@ -2,6 +2,7 @@
 # SPDX-License-Identifier: Apache-2.0
 
 import logging
+import warnings
 from typing import Optional
 
 import numpy as np
@@ -26,6 +27,7 @@ def _require_ffcv():
 
 
 def ffcv_monkey_patches():
+    warnings.warn(DeprecationWarning('ffcv_monkey_patches is deprecated and will be removed in v0.18'))
     _require_ffcv()
 
     # ffcv's __len__ function is expensive as it always calls self.next_traversal_order which does shuffling.
@@ -62,6 +64,7 @@ def write_ffcv_dataset(dataset: Optional[Dataset] = None,
         jpeg_quality (float): Quality to use for jpeg compression. Default: ``90``.
         chunk_size (int): Size of chunks processed by each worker during conversion. Default: ``100``.
     """
+    warnings.warn(DeprecationWarning('write_ffcv_dataset is deprecated and will be removed in v0.18'))
 
     _require_ffcv()
     if dataset is None:

composer/datasets/imagenet.py

@@ -8,6 +8,7 @@
 """
 
 import os
+import warnings
 from typing import Any, Dict, List, Optional
 
 import numpy as np
@@ -57,6 +58,8 @@ def build_imagenet_dataloader(
         crop size (int): The crop size to use. Default: ``224``.
         **dataloader_kwargs (Dict[str, Any]): Additional settings for the dataloader (e.g. num_workers, etc.)
     """
+    warnings.warn(DeprecationWarning('build_imagenet_dataloader is deprecated and will be removed in v0.18'))
+
     if global_batch_size % dist.get_world_size() != 0:
         raise ValueError(
             f'global_batch_size ({global_batch_size}) must be divisible by world_size ({dist.get_world_size()}).')
@@ -126,6 +129,8 @@ def build_synthetic_imagenet_dataloader(
         shuffle (bool): whether to shuffle the dataset. Default: ``True``.
         **dataloader_kwargs (Dict[str, Any]): Additional settings for the dataloader (e.g. num_workers, etc.)
     """
+    warnings.warn(DeprecationWarning('build_synthetic_imagenet_dataloader is deprecated and will be removed in v0.18'))
+
     if global_batch_size % dist.get_world_size() != 0:
         raise ValueError(
             f'global_batch_size ({global_batch_size}) must be divisible by world_size ({dist.get_world_size()}).')
@@ -166,6 +171,7 @@ def write_ffcv_imagenet(
         split (str): 'train' or 'val'. Default: ``train``.
         num_workers (int): Number of workers to use for conversion. Default: ``8``.
     """
+    warnings.warn(DeprecationWarning('write_ffcv_imagenet is deprecated and will be removed in v0.18'))
 
     if dist.get_local_rank() == 0:
         ds = ImageFolder(os.path.join(datadir, split))
@@ -205,6 +211,8 @@ def build_ffcv_imagenet_dataloader(
         prefetch_factor (int): Number of batches to prefect. Default: ``2``.
         num_workers (int): Number of workers. Default: ``8``.
     """
+    warnings.warn(DeprecationWarning('build_ffcv_imagenet_dataloader is deprecated and will be removed in v0.18'))
+
     try:
         import ffcv
         from ffcv.fields.decoders import CenterCropRGBImageDecoder, IntDecoder, RandomResizedCropRGBImageDecoder
@@ -317,6 +325,9 @@ def build_streaming_imagenet1k_dataloader(
             Defaults to ``None``, which is interpreted as the number of nodes of the initial run.
         **dataloader_kwargs (Dict[str, Any]): Additional settings for the dataloader (e.g. num_workers, etc.)
     """
+    warnings.warn(
+        DeprecationWarning('build_streaming_imagenet1k_dataloader is deprecated and will be removed in v0.18'))
+
     if global_batch_size % dist.get_world_size() != 0:
         raise ValueError(
             f'global_batch_size ({global_batch_size}) must be divisible by world_size ({dist.get_world_size()}).')

composer/datasets/lm_dataset.py

@@ -2,6 +2,7 @@
 # SPDX-License-Identifier: Apache-2.0
 
 import logging
+import warnings
 from typing import List, cast
 
 from torch.utils.data import DataLoader, Dataset
@@ -48,6 +49,8 @@ def build_lm_dataloader(
             ``1.0``.
         **dataloader_kwargs (Dict[str, Any]): Additional settings for the dataloader (e.g. num_workers, etc.)
     """
+    warnings.warn(DeprecationWarning('build_lm_dataloader is deprecated and will be removed in v0.18'))
+
     if global_batch_size % dist.get_world_size() != 0:
         raise ValueError(
             f'global_batch_size ({global_batch_size}) must be divisible by world_size ({dist.get_world_size()}).')

composer/datasets/mnist.py

@@ -1,6 +1,7 @@
 # Copyright 2022 MosaicML Composer authors
 # SPDX-License-Identifier: Apache-2.0
 
+import warnings
 from typing import Any
 
 from torch.utils.data import DataLoader
@@ -33,6 +34,8 @@ def build_mnist_dataloader(
         shuffle (bool): Shuffle the dataset. Default: ``True``.
         **dataloader_kwargs (Any): Additional settings for the dataloader (e.g. num_workers, etc.)
     """
+    warnings.warn(DeprecationWarning('build_mnist_dataloader is deprecated and will be removed in v0.18'))
+
     if global_batch_size % dist.get_world_size() != 0:
         raise ValueError(
             f'global_batch_size ({global_batch_size}) must be divisible by world_size ({dist.get_world_size()}).')

composer/datasets/synthetic.py

@@ -5,6 +5,7 @@
 
 from __future__ import annotations
 
+import warnings
 from typing import Callable, Optional, Sequence, Union
 
 import torch
@@ -79,6 +80,8 @@ def __init__(self,
                  device: str = 'cpu',
                  memory_format: Union[str, MemoryFormat] = MemoryFormat.CONTIGUOUS_FORMAT,
                  transform: Optional[Callable] = None):
+        warnings.warn(DeprecationWarning('SyntheticBatchPairDataset is deprecated and will be removed in v0.18'))
+
         self.total_dataset_size = total_dataset_size
         self.data_shape = data_shape
         self.num_unique_samples_to_create = num_unique_samples_to_create
@@ -187,6 +190,7 @@ def __init__(self,
                  num_classes: Optional[int] = None,
                  label_shape: Optional[Sequence[int]] = None,
                  transform: Optional[Callable] = None):
+        warnings.warn(DeprecationWarning('SyntheticPILDataset is deprecated and will be removed in v0.18'))
         super().__init__(root='', transform=transform)
         self._dataset = SyntheticBatchPairDataset(
             total_dataset_size=total_dataset_size,

composer/datasets/utils.py

@@ -5,6 +5,7 @@
 
 import logging
 import textwrap
+import warnings
 from typing import Callable, List, Tuple, Union
 
 import numpy as np
@@ -44,6 +45,7 @@ def __init__(self,
                  mean: Tuple[float, float, float],
                  std: Tuple[float, float, float],
                  ignore_background: bool = False):
+        warnings.warn(DeprecationWarning('NormalizationFn is deprecated and will be removed in v0.18'))
         self.mean = mean
         self.std = std
         self.ignore_background = ignore_background
@@ -87,6 +89,7 @@ def pil_image_collate(
         (torch.Tensor, torch.Tensor): Tuple of (image tensor, target tensor)
             The image tensor will be four-dimensional (NCHW or NHWC, depending on the ``memory_format``).
     """
+    warnings.warn(DeprecationWarning('pil_image_collate is deprecated and will be removed in v0.18'))
     imgs = [sample[0] for sample in batch]
     w, h = imgs[0].size
     image_tensor = torch.zeros((len(imgs), 3, h, w), dtype=torch.uint8).contiguous(memory_format=memory_format)
@@ -137,6 +140,7 @@ def add_vision_dataset_transform(dataset: VisionDataset, transform: Callable, is
     Returns:
         None: The ``dataset`` is modified in-place.
     """
+    warnings.warn(DeprecationWarning('add_vision_dataset_transform is deprecated and will be removed in v0.18'))
 
     transform_added_logstring = textwrap.dedent(f"""\
         Transform {transform} added to dataset.

composer/models/bert/model.py

@@ -5,6 +5,7 @@
 
 from __future__ import annotations
 
+import warnings
 from typing import Optional
 
 from torchmetrics import MeanSquaredError
@@ -70,6 +71,8 @@ def create_bert_mlm(use_pretrained: Optional[bool] = False,
         model = create_bert_mlm()
 
     """
+    warnings.warn(DeprecationWarning('create_bert_mlm is deprecated and will be removed in v0.18'))
+
     try:
         import transformers
     except ImportError as e:
@@ -175,6 +178,8 @@ def create_bert_classification(num_labels: int = 2,
         For the classification case (when ``num_labels > 1``), the training loss is :class:`~torch.nn.CrossEntropyLoss`, and the train/validation
         metrics are :class:`~torchmetrics.MulticlassAccuracy` and :class:`~torchmetrics.MatthewsCorrCoef`, as well as :class:`.BinaryF1Score` if ``num_labels == 2``.
     """
+    warnings.warn(DeprecationWarning('create_bert_classification is deprecated and will be removed in v0.18'))
+
     try:
         import transformers
     except ImportError as e:

composer/models/classify_mnist/model.py

@@ -3,6 +3,7 @@
 
 """A simple convolutional neural network extending :class:`.ComposerClassifier`."""
 
+import warnings
 from typing import List, Optional, Sequence, Union
 
 import torch
@@ -19,6 +20,7 @@ class Model(nn.Module):
     """Toy convolutional neural network architecture in pytorch for MNIST."""
 
     def __init__(self, initializers: Sequence[Union[str, Initializer]], num_classes: int = 10):
+        warnings.warn(DeprecationWarning('Model is deprecated and will be removed in v0.18'))
         super().__init__()
 
         self.num_classes = num_classes
@@ -65,6 +67,7 @@ def mnist_model(num_classes: int = 10, initializers: Optional[List[Initializer]]
 
         model = mnist_model()
     """
+    warnings.warn(DeprecationWarning('mnist_model is deprecated and will be removed in v0.18'))
 
     if initializers is None:
         initializers = []