|
|
|
|
|
|
|
|
|
""" FDViT model configuration"""
|
|
|
|
from collections import OrderedDict
|
|
from typing import Mapping
|
|
|
|
from packaging import version
|
|
|
|
from transformers.configuration_utils import PretrainedConfig
|
|
from transformers.onnx import OnnxConfig
|
|
from transformers.utils import logging
|
|
|
|
|
|
logger = logging.get_logger(__name__)
|
|
|
|
|
|
|
|
|
|
|
|
class FDViTConfig(PretrainedConfig):
|
|
r"""
|
|
This is the configuration class to store the configuration of a [`FDViTModel`]. It is used to instantiate an FDViT
|
|
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 FDViT
|
|
[amd/fdvit_ti](https://huggingface.co/amd/fdvit_ti) architecture.
|
|
|
|
Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
|
|
documentation from [`PretrainedConfig`] for more information.
|
|
|
|
|
|
Args:
|
|
image_size (`int`, *optional*, defaults to 224):
|
|
The size (resolution) of each image.
|
|
patch_size (`int`, *optional*, defaults to 16):
|
|
The size of the input patch.
|
|
stride (`int`, *optional*, defaults to 16):
|
|
The stride of the input patch.
|
|
base_dims (`list`, *optional*, defaults to `[32, 23, 21, 23, 26]`):
|
|
The basic dimension of each encoder block.
|
|
depth (`list`, *optional*, defaults to `[2, 3, 3, 2, 2]`):
|
|
The depth of each encoder block.
|
|
heads (`list`, *optional*, defaults to `[2, 4, 6, 8, 10]`):
|
|
The depth of each encoder block.
|
|
channels (`list`, *optional*, defaults to `[64, 92, 126, 184, 260]`):
|
|
The depth of each encoder block.
|
|
out_size (`list`, *optional*, defaults to `[27, 19, 14, 10, 7]`):
|
|
The output size of each encoder block.
|
|
mlp_ratio (`float`, *optional*, defaults to 4.0):
|
|
The ratio of the number of channels in the output of the MLP to the number of channels in the input.
|
|
num_classes (`int`, *optional*, defaults to 1000):
|
|
The number of classes of the dataset.
|
|
in_chans (`int`, *optional*, defaults to 3):
|
|
The number of channels in the input image.
|
|
attn_drop_rate (`float`, *optional*, defaults to 0.0):
|
|
The attention drop rate for the attention dropout layers.
|
|
drop_rate (`float`, *optional*, defaults to 0.0):
|
|
The dropout rate for the dropout layers.
|
|
drop_path_rate (`float`, *optional*, defaults to 0.1):
|
|
The droppath rate for the droppath layers.
|
|
initializer_range (`float`, *optional*, defaults to 0.02):
|
|
The initializer range for the weights.
|
|
|
|
|
|
Example:
|
|
|
|
```python
|
|
>>> from transformers import FDViTConfig, FDViTModel
|
|
|
|
>>> # Initializing a FDViT fdvit_ti style configuration
|
|
>>> configuration = FDViTConfig()
|
|
|
|
>>> # Initializing a model (with random weights) from the fdvit_ti style configuration
|
|
>>> model = FDViTModel(configuration)
|
|
|
|
>>> # Accessing the model configuration
|
|
>>> configuration = model.config
|
|
```"""
|
|
|
|
model_type = "fdvit"
|
|
|
|
def __init__(
|
|
self,
|
|
image_size=224,
|
|
patch_size=16,
|
|
stride=8,
|
|
base_dims=[32, 23, 21, 23, 26],
|
|
depth=[2, 3, 3, 2, 2],
|
|
heads=[2, 4, 6, 8, 10],
|
|
channels=[64,92,126,184,260],
|
|
out_size=[27, 19, 14, 10, 7],
|
|
mlp_ratio=4,
|
|
num_classes=1000,
|
|
in_chans=3,
|
|
attn_drop_rate=0.0,
|
|
drop_rate=0.0,
|
|
drop_path_rate=0.1,
|
|
initializer_range=0.02,
|
|
**kwargs,
|
|
):
|
|
super().__init__(**kwargs)
|
|
|
|
self.image_size = image_size
|
|
self.patch_size = patch_size
|
|
self.stride = stride
|
|
self.base_dims = base_dims
|
|
self.depth = depth
|
|
self.heads = heads
|
|
self.channels = channels
|
|
self.out_size = out_size
|
|
self.mlp_ratio = mlp_ratio
|
|
self.num_classes = num_classes
|
|
self.in_chans = in_chans
|
|
self.attn_drop_rate = attn_drop_rate
|
|
self.drop_rate = drop_rate
|
|
self.drop_path_rate = drop_path_rate
|
|
self.initializer_range = initializer_range
|
|
|
|
|
|
|
|
class FDViTOnnxConfig(OnnxConfig):
|
|
torch_onnx_minimum_version = version.parse("1.11")
|
|
|
|
@property
|
|
def inputs(self) -> Mapping[str, Mapping[int, str]]:
|
|
return OrderedDict(
|
|
[
|
|
("pixel_values", {0: "batch", 1: "num_channels", 2: "height", 3: "width"}),
|
|
]
|
|
)
|
|
|
|
@property
|
|
def atol_for_validation(self) -> float:
|
|
return 1e-4 |
