SwiftFormer
Overview
The SwiftFormer model was proposed in SwiftFormer: Efficient Additive Attention for Transformer-based Real-time Mobile Vision Applications by Abdelrahman Shaker, Muhammad Maaz, Hanoona Rasheed, Salman Khan, Ming-Hsuan Yang, Fahad Shahbaz Khan.
The SwiftFormer paper introduces a novel efficient additive attention mechanism that effectively replaces the quadratic matrix multiplication operations in the self-attention computation with linear element-wise multiplications. A series of models called ‘SwiftFormer’ is built based on this, which achieves state-of-the-art performance in terms of both accuracy and mobile inference speed. Even their small variant achieves 78.5% top-1 ImageNet1K accuracy with only 0.8 ms latency on iPhone 14, which is more accurate and 2× faster compared to MobileViT-v2.
The abstract from the paper is the following:
Self-attention has become a defacto choice for capturing global context in various vision applications. However, its quadratic computational complexity with respect to image resolution limits its use in real-time applications, especially for deployment on resource-constrained mobile devices. Although hybrid approaches have been proposed to combine the advantages of convolutions and self-attention for a better speed-accuracy trade-off, the expensive matrix multiplication operations in self-attention remain a bottleneck. In this work, we introduce a novel efficient additive attention mechanism that effectively replaces the quadratic matrix multiplication operations with linear element-wise multiplications. Our design shows that the key-value interaction can be replaced with a linear layer without sacrificing any accuracy. Unlike previous state-of-the-art methods, our efficient formulation of self-attention enables its usage at all stages of the network. Using our proposed efficient additive attention, we build a series of models called “SwiftFormer” which achieves state-of-the-art performance in terms of both accuracy and mobile inference speed. Our small variant achieves 78.5% top-1 ImageNet-1K accuracy with only 0.8 ms latency on iPhone 14, which is more accurate and 2x faster compared to MobileViT-v2.
This model was contributed by shehan97. The TensorFlow version was contributed by joaocmd. The original code can be found here.
SwiftFormerConfig
class transformers.SwiftFormerConfig
< source >( image_size = 224 num_channels = 3 depths = [3, 3, 6, 4] embed_dims = [48, 56, 112, 220] mlp_ratio = 4 downsamples = [True, True, True, True] hidden_act = 'gelu' down_patch_size = 3 down_stride = 2 down_pad = 1 drop_path_rate = 0.0 drop_mlp_rate = 0.0 drop_conv_encoder_rate = 0.0 use_layer_scale = True layer_scale_init_value = 1e-05 batch_norm_eps = 1e-05 **kwargs )
Parameters
- image_size (
int, optional, defaults to 224) — The size (resolution) of each image - num_channels (
int, optional, defaults to 3) — The number of input channels - depths (
List[int], optional, defaults to[3, 3, 6, 4]) — Depth of each stage - embed_dims (
List[int], optional, defaults to[48, 56, 112, 220]) — The embedding dimension at each stage - mlp_ratio (
int, optional, defaults to 4) — Ratio of size of the hidden dimensionality of an MLP to the dimensionality of its input. - downsamples (
List[bool], optional, defaults to[True, True, True, True]) — Whether or not to downsample inputs between two stages. - hidden_act (
str, optional, defaults to"gelu") — The non-linear activation function (string)."gelu","relu","selu"and"gelu_new"are supported. - down_patch_size (
int, optional, defaults to 3) — The size of patches in downsampling layers. - down_stride (
int, optional, defaults to 2) — The stride of convolution kernels in downsampling layers. - down_pad (
int, optional, defaults to 1) — Padding in downsampling layers. - drop_path_rate (
float, optional, defaults to 0.0) — Rate at which to increase dropout probability in DropPath. - drop_mlp_rate (
float, optional, defaults to 0.0) — Dropout rate for the MLP component of SwiftFormer. - drop_conv_encoder_rate (
float, optional, defaults to 0.0) — Dropout rate for the ConvEncoder component of SwiftFormer. - use_layer_scale (
bool, optional, defaults toTrue) — Whether to scale outputs from token mixers. - layer_scale_init_value (
float, optional, defaults to 1e-05) — Factor by which outputs from token mixers are scaled. - batch_norm_eps (
float, optional, defaults to 1e-05) — The epsilon used by the batch normalization layers.
This is the configuration class to store the configuration of a SwiftFormerModel. It is used to instantiate an SwiftFormer model according to the specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of the SwiftFormer MBZUAI/swiftformer-xs architecture.
Configuration objects inherit from PretrainedConfig and can be used to control the model outputs. Read the documentation from PretrainedConfig for more information.
Example:
>>> from transformers import SwiftFormerConfig, SwiftFormerModel
>>> # Initializing a SwiftFormer swiftformer-base-patch16-224 style configuration
>>> configuration = SwiftFormerConfig()
>>> # Initializing a model (with random weights) from the swiftformer-base-patch16-224 style configuration
>>> model = SwiftFormerModel(configuration)
>>> # Accessing the model configuration
>>> configuration = model.configSwiftFormerModel
class transformers.SwiftFormerModel
< source >( config: SwiftFormerConfig )
Parameters
- config (SwiftFormerConfig) — Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.
The bare SwiftFormer Model transformer outputting raw hidden-states without any specific head on top. This model is a PyTorch torch.nn.Module subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
forward
< source >( pixel_values: typing.Optional[torch.Tensor] = None output_hidden_states: typing.Optional[bool] = None return_dict: typing.Optional[bool] = None ) → transformers.modeling_outputs.BaseModelOutputWithNoAttention or tuple(torch.FloatTensor)
Parameters
- pixel_values (
torch.FloatTensorof shape(batch_size, num_channels, height, width)) — Pixel values. Pixel values can be obtained using AutoImageProcessor. See ViTImageProcessor.call() for details. - output_hidden_states (
bool, optional) — Whether or not to return the hidden states of all layers. Seehidden_statesunder returned tensors for more detail. - return_dict (
bool, optional) — Whether or not to return a ModelOutput instead of a plain tuple.
Returns
transformers.modeling_outputs.BaseModelOutputWithNoAttention or tuple(torch.FloatTensor)
A transformers.modeling_outputs.BaseModelOutputWithNoAttention or a tuple of
torch.FloatTensor (if return_dict=False is passed or when config.return_dict=False) comprising various
elements depending on the configuration (SwiftFormerConfig) and inputs.
-
last_hidden_state (
torch.FloatTensorof shape(batch_size, num_channels, height, width)) — Sequence of hidden-states at the output of the last layer of the model. -
hidden_states (
tuple(torch.FloatTensor), optional, returned whenoutput_hidden_states=Trueis passed or whenconfig.output_hidden_states=True) — Tuple oftorch.FloatTensor(one for the output of the embeddings, if the model has an embedding layer, + one for the output of each layer) of shape(batch_size, num_channels, height, width).Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
The SwiftFormerModel forward method, overrides the __call__ special method.
Although the recipe for forward pass needs to be defined within this function, one should call the Module
instance afterwards instead of this since the former takes care of running the pre and post processing steps while
the latter silently ignores them.
Example:
>>> from transformers import AutoImageProcessor, SwiftFormerModel
>>> import torch
>>> from datasets import load_dataset
>>> dataset = load_dataset("huggingface/cats-image", trust_remote_code=True)
>>> image = dataset["test"]["image"][0]
>>> image_processor = AutoImageProcessor.from_pretrained("MBZUAI/swiftformer-xs")
>>> model = SwiftFormerModel.from_pretrained("MBZUAI/swiftformer-xs")
>>> inputs = image_processor(image, return_tensors="pt")
>>> with torch.no_grad():
... outputs = model(**inputs)
>>> last_hidden_states = outputs.last_hidden_state
>>> list(last_hidden_states.shape)
[1, 220, 7, 7]SwiftFormerForImageClassification
class transformers.SwiftFormerForImageClassification
< source >( config: SwiftFormerConfig )
Parameters
- config (SwiftFormerConfig) — Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.
SwiftFormer Model transformer with an image classification head on top (e.g. for ImageNet).
This model is a PyTorch torch.nn.Module subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
forward
< source >( pixel_values: typing.Optional[torch.Tensor] = None labels: typing.Optional[torch.Tensor] = None output_hidden_states: typing.Optional[bool] = None return_dict: typing.Optional[bool] = None ) → transformers.modeling_outputs.ImageClassifierOutputWithNoAttention or tuple(torch.FloatTensor)
Parameters
- pixel_values (
torch.FloatTensorof shape(batch_size, num_channels, height, width)) — Pixel values. Pixel values can be obtained using AutoImageProcessor. See ViTImageProcessor.call() for details. - output_hidden_states (
bool, optional) — Whether or not to return the hidden states of all layers. Seehidden_statesunder returned tensors for more detail. - return_dict (
bool, optional) — Whether or not to return a ModelOutput instead of a plain tuple. - labels (
torch.LongTensorof shape(batch_size,), optional) — Labels for computing the image classification/regression loss. Indices should be in[0, ..., config.num_labels - 1]. Ifconfig.num_labels == 1a regression loss is computed (Mean-Square loss), Ifconfig.num_labels > 1a classification loss is computed (Cross-Entropy).
Returns
transformers.modeling_outputs.ImageClassifierOutputWithNoAttention or tuple(torch.FloatTensor)
A transformers.modeling_outputs.ImageClassifierOutputWithNoAttention or a tuple of
torch.FloatTensor (if return_dict=False is passed or when config.return_dict=False) comprising various
elements depending on the configuration (SwiftFormerConfig) and inputs.
- loss (
torch.FloatTensorof shape(1,), optional, returned whenlabelsis provided) — Classification (or regression if config.num_labels==1) loss. - logits (
torch.FloatTensorof shape(batch_size, config.num_labels)) — Classification (or regression if config.num_labels==1) scores (before SoftMax). - hidden_states (
tuple(torch.FloatTensor), optional, returned whenoutput_hidden_states=Trueis passed or whenconfig.output_hidden_states=True) — Tuple oftorch.FloatTensor(one for the output of the embeddings, if the model has an embedding layer, + one for the output of each stage) of shape(batch_size, num_channels, height, width). Hidden-states (also called feature maps) of the model at the output of each stage.
The SwiftFormerForImageClassification forward method, overrides the __call__ special method.
Although the recipe for forward pass needs to be defined within this function, one should call the Module
instance afterwards instead of this since the former takes care of running the pre and post processing steps while
the latter silently ignores them.
Example:
>>> from transformers import AutoImageProcessor, SwiftFormerForImageClassification
>>> import torch
>>> from datasets import load_dataset
>>> dataset = load_dataset("huggingface/cats-image", trust_remote_code=True)
>>> image = dataset["test"]["image"][0]
>>> image_processor = AutoImageProcessor.from_pretrained("MBZUAI/swiftformer-xs")
>>> model = SwiftFormerForImageClassification.from_pretrained("MBZUAI/swiftformer-xs")
>>> inputs = image_processor(image, return_tensors="pt")
>>> with torch.no_grad():
... logits = model(**inputs).logits
>>> # model predicts one of the 1000 ImageNet classes
>>> predicted_label = logits.argmax(-1).item()
>>> print(model.config.id2label[predicted_label])
tabby, tabby catTFSwiftFormerModel
class transformers.TFSwiftFormerModel
< source >( config: SwiftFormerConfig *inputs **kwargs )
Parameters
- config (SwiftFormerConfig) — Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.
The bare TFSwiftFormer Model transformer outputting raw hidden-states without any specific head on top. This model inherits from TFPreTrainedModel. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads etc.)
This model is also a keras.Model subclass. Use it as a regular TF 2.0 Keras Model and refer to the TF 2.0 documentation for all matter related to general usage and behavior.
TF 2.0 models accepts two formats as inputs:
- having all inputs as keyword arguments (like PyTorch models), or
- having all inputs as a list, tuple or dict in the first positional arguments.
This second option is useful when using
keras.Model.fitmethod which currently requires having all the tensors in the first argument of the model call function:model(inputs). If you choose this second option, there are three possibilities you can use to gather all the input Tensors in the first positional argument : - a single Tensor with
input_idsonly and nothing else:model(input_ids) - a list of varying length with one or several input Tensors IN THE ORDER given in the docstring:
model([input_ids, attention_mask])ormodel([input_ids, attention_mask, token_type_ids]) - a dictionary with one or several input Tensors associated to the input names given in the docstring:
model({"input_ids": input_ids, "token_type_ids": token_type_ids})
call
< source >( pixel_values: typing.Optional[tensorflow.python.framework.tensor.Tensor] = None output_hidden_states: typing.Optional[bool] = None return_dict: typing.Optional[bool] = None training: bool = False )
Parameters
- pixel_values (
tf.Tensorof shape(batch_size, num_channels, height, width)) — Pixel values. Pixel values can be obtained using AutoImageProcessor. See ViTImageProcessor.call() for details. - output_hidden_states (
bool, optional) — Whether or not to return the hidden states of all layers. Seehidden_statesunder returned tensors for more detail. - return_dict (
bool, optional) — Whether or not to return a ModelOutput instead of a plain tuple. - training (
bool, optional, defaults toFalse) — Whether or not to run the model in training mode.
The TFSwiftFormerModel forward method, overrides the __call__ special method.
Although the recipe for forward pass needs to be defined within this function, one should call the Module
instance afterwards instead of this since the former takes care of running the pre and post processing steps while
the latter silently ignores them.
TFSwiftFormerForImageClassification
class transformers.TFSwiftFormerForImageClassification
< source >( config: SwiftFormerConfig **kwargs )
Parameters
- config (SwiftFormerConfig) — Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.
TFSwiftFormer Model transformer with an image classification head on top (e.g. for ImageNet).
This model inherits from TFPreTrainedModel. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads etc.)
This model is also a keras.Model subclass. Use it as a regular TF 2.0 Keras Model and refer to the TF 2.0 documentation for all matter related to general usage and behavior.
TF 2.0 models accepts two formats as inputs:
- having all inputs as keyword arguments (like PyTorch models), or
- having all inputs as a list, tuple or dict in the first positional arguments.
This second option is useful when using
keras.Model.fitmethod which currently requires having all the tensors in the first argument of the model call function:model(inputs). If you choose this second option, there are three possibilities you can use to gather all the input Tensors in the first positional argument : - a single Tensor with
input_idsonly and nothing else:model(input_ids) - a list of varying length with one or several input Tensors IN THE ORDER given in the docstring:
model([input_ids, attention_mask])ormodel([input_ids, attention_mask, token_type_ids]) - a dictionary with one or several input Tensors associated to the input names given in the docstring:
model({"input_ids": input_ids, "token_type_ids": token_type_ids})
call
< source >( pixel_values: typing.Optional[tensorflow.python.framework.tensor.Tensor] = None labels: typing.Optional[tensorflow.python.framework.tensor.Tensor] = None output_hidden_states: typing.Optional[bool] = None return_dict: typing.Optional[bool] = None training: bool = False )
Parameters
- pixel_values (
tf.Tensorof shape(batch_size, num_channels, height, width)) — Pixel values. Pixel values can be obtained using AutoImageProcessor. See ViTImageProcessor.call() for details. - output_hidden_states (
bool, optional) — Whether or not to return the hidden states of all layers. Seehidden_statesunder returned tensors for more detail. - return_dict (
bool, optional) — Whether or not to return a ModelOutput instead of a plain tuple. - training (
bool, optional, defaults toFalse) — Whether or not to run the model in training mode. - labels (
tf.Tensorof shape(batch_size,), optional) — Labels for computing the image classification/regression loss. Indices should be in[0, ..., config.num_labels - 1]. Ifconfig.num_labels == 1a regression loss is computed (Mean-Square loss), Ifconfig.num_labels > 1a classification loss is computed (Cross-Entropy).
The TFSwiftFormerForImageClassification forward method, overrides the __call__ special method.
Although the recipe for forward pass needs to be defined within this function, one should call the Module
instance afterwards instead of this since the former takes care of running the pre and post processing steps while
the latter silently ignores them.