#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
#           This file was automatically generated from src/transformers/models/rf_detr/modular_rf_detr.py.
#               Do NOT edit this file manually as any edits will be overwritten by the generation of
#             the file from the modular. If any change should be done, please apply the change to the
#                          modular_rf_detr.py file directly. One of our CI enforces this.
#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from huggingface_hub.dataclasses import strict

from ...backbone_utils import BackboneConfigMixin, consolidate_backbone_kwargs_to_config
from ...configuration_utils import PreTrainedConfig
from ...utils import auto_docstring
from ..auto import AutoConfig


@auto_docstring(checkpoint="Roboflow/rf-detr-base")
@strict
class RfDetrDinov2Config(BackboneConfigMixin, PreTrainedConfig):
    r"""
    layerscale_value (`float`, *optional*, defaults to 1.0):
        Initial value to use for layer scale.
    drop_path_rate (`float`, *optional*, defaults to 0.0):
        Stochastic depth rate per sample (when applied in the main path of residual layers).
    use_swiglu_ffn (`bool`, *optional*, defaults to `False`):
        Whether to use the SwiGLU feedforward neural network.
    apply_layernorm (`bool`, *optional*, defaults to `True`):
        Whether to apply layer normalization to the feature maps in case the model is used as backbone.
    reshape_hidden_states (`bool`, *optional*, defaults to `True`):
        Whether to reshape the feature maps to 4D tensors of shape `(batch_size, d_model, height, width)` in
        case the model is used as backbone. If `False`, the feature maps will be 3D tensors of shape `(batch_size,
        seq_len, d_model)`.
    use_mask_token (`bool`, *optional*, defaults to `True`):
        Whether to use mask_token in embeddings.
    num_windows (`int`, *optional*, defaults to 4):
        Number of windows to use for windowed attention. If 1, no windowed attention is used.

    Example:

    ```python
    >>> from transformers import RfDetrDinov2Config, RfDetrDinov2Backbone

    >>> # Initializing a RfDetrDinov2 base style configuration
    >>> configuration = RfDetrDinov2Config()

    >>> # Initializing a model (with random weights) from the base style configuration
    >>> model = RfDetrDinov2Backbone(configuration)

    >>> # Accessing the model configuration
    >>> configuration = model.config
    ```"""

    model_type = "rf_detr_dinov2"

    hidden_size: int = 768
    num_hidden_layers: int = 12
    num_attention_heads: int = 12
    mlp_ratio: int = 4
    hidden_act: str = "gelu"
    hidden_dropout_prob: float | int = 0.0
    attention_probs_dropout_prob: float | int = 0.0
    initializer_range: float = 0.02
    layer_norm_eps: float = 1e-6
    image_size: int | list[int] | tuple[int, int] = 224
    patch_size: int | list[int] | tuple[int, int] = 14
    num_channels: int = 3
    qkv_bias: bool = True
    layerscale_value: float = 1.0
    drop_path_rate: float | int = 0.0
    use_swiglu_ffn: bool = False
    _out_features: list[str] | None = None
    _out_indices: list[int] | None = None
    apply_layernorm: bool = True
    reshape_hidden_states: bool = True
    use_mask_token: bool = True

    num_windows: int = 4

    def __post_init__(self, **kwargs):
        self.stage_names = ["stem"] + [f"stage{idx}" for idx in range(1, self.num_hidden_layers + 1)]
        self.set_output_features_output_indices(
            out_indices=kwargs.pop("out_indices", None), out_features=kwargs.pop("out_features", None)
        )
        window_block_indexes = set(range(self._out_indices[-1] + 1))
        window_block_indexes.difference_update(self._out_indices)
        self.window_block_indexes = list(window_block_indexes)
        super().__post_init__(**kwargs)


@auto_docstring(checkpoint="Roboflow/rf-detr-base")
@strict
class RfDetrConfig(PreTrainedConfig):
    r"""
    hidden_expansion (`float`, *optional*, defaults to 0.5):
        Expansion factor for hidden dimensions in the projector layers.
    c2f_num_blocks (`int`, *optional*, defaults to 3):
        Number of blocks in the C2F layer.
    activation_function (`str`, *optional*, defaults to `"silu"`):
        The non-linear activation function in the projector. Supported values are `"silu"`, `"relu"`, `"gelu"`.
    decoder_n_points (`int`, *optional*, defaults to 4):
        The number of sampled keys in each feature level for each attention head in the decoder.
    decoder_layers (`int`, *optional*, defaults to 3):
        Number of decoder layers in the transformer.
    decoder_self_attention_heads (`int`, *optional*, defaults to 8):
        Number of attention heads for each attention layer in the decoder self-attention.
    decoder_cross_attention_heads (`int`, *optional*, defaults to 16):
        Number of attention heads for each attention layer in the decoder cross-attention.
    decoder_activation_function (`str`, *optional*, defaults to `"relu"`):
        The non-linear activation function in the decoder. Supported values are `"relu"`, `"silu"`, `"gelu"`.
    num_queries (`int`, *optional*, defaults to 300):
        Number of object queries, i.e. detection slots. This is the maximal number of objects
        [`RfDetrModel`] can detect in a single image.
    group_detr (`int`, *optional*, defaults to 13):
        Number of groups for Group DETR attention mechanism, which helps reduce computational complexity.
    disable_custom_kernels (`bool`, *optional*, defaults to `True`):
        Disable the use of custom CUDA and CPU kernels. This option is necessary for the ONNX export, as custom
        kernels are not supported by PyTorch ONNX export.
    class_loss_coefficient (`float`, *optional*, defaults to 1):
        Relative weight of the classification loss in the Hungarian matching cost.
    dice_loss_coefficient (`float`, *optional*, defaults to 1):
        Relative weight of the DICE/F-1 loss in the object detection loss.
    bbox_loss_coefficient (`float`, *optional*, defaults to 5):
        Relative weight of the L1 bounding box loss in the object detection loss.
    giou_loss_coefficient (`float`, *optional*, defaults to 2):
        Relative weight of the generalized IoU loss in the object detection loss.
    num_feature_levels (`int`, *optional*, defaults to 1):
        Number of feature levels used in the multiscale deformable attention.
    mask_loss_coefficient (`float`, *optional*, defaults to 1):
        Relative weight of the Focal loss in the instance segmentation mask loss.
    mask_point_sample_ratio (`int`, *optional*, defaults to 16):
        The ratio of points to sample for the mask loss calculation.
    mask_downsample_ratio (`int`, *optional*, defaults to 4):
        The downsample ratio for the segmentation masks compared to the input image resolution.
    mask_class_loss_coefficient (`float`, *optional*, defaults to 5.0):
        Relative weight of the Focal loss in the instance segmentation loss.
    mask_dice_loss_coefficient (`float`, *optional*, defaults to 5.0):
        Relative weight of the DICE/F-1 loss in the instance segmentation loss.
    segmentation_head_activation_function (`str`, *optional*, defaults to `"gelu"`):
        The non-linear activation function in the segmentation head. Supported values are `"relu"`, `"silu"`, `"gelu"`.

    Examples:

    ```python
    >>> from transformers import RfDetrConfig, RfDetrModel

    >>> # Initializing a RF-DETR roboflow/rf-detr-base style configuration
    >>> configuration = RfDetrConfig()

    >>> # Initializing a model (with random weights) from the Roboflow/rf-detr-base style configuration
    >>> model = RfDetrModel(configuration)

    >>> # Accessing the model configuration
    >>> configuration = model.config
    ```"""

    model_type = "rf_detr"
    sub_configs = {"backbone_config": AutoConfig}

    backbone_config: dict | PreTrainedConfig | None = None
    hidden_expansion: float = 0.5
    c2f_num_blocks: int = 3
    activation_function: str = "silu"
    dropout: float = 0.1
    decoder_ffn_dim: int = 2048
    decoder_n_points: int = 4
    decoder_layers: int = 3
    decoder_self_attention_heads: int = 8
    decoder_cross_attention_heads: int = 16
    decoder_activation_function: str = "relu"
    num_queries: int = 300
    attention_bias: bool = True
    attention_dropout: float | int = 0.0
    activation_dropout: float | int = 0.0
    group_detr: int = 13
    init_std: float = 0.02
    disable_custom_kernels: bool = True
    class_cost: int | float = 2
    bbox_cost: int | float = 5
    giou_cost: int | float = 2
    class_loss_coefficient: int | float = 1
    dice_loss_coefficient: int | float = 1
    bbox_loss_coefficient: int | float = 5
    giou_loss_coefficient: int | float = 2
    eos_coefficient: float = 0.1
    focal_alpha: float = 0.25
    auxiliary_loss: bool = True
    d_model: int = 256

    layer_norm_eps: float = 1e-5
    num_feature_levels: int = 1
    mask_loss_coefficient: int | float = 1
    mask_point_sample_ratio: int = 16
    mask_downsample_ratio: int = 4
    mask_class_loss_coefficient: int | float = 5.0
    mask_dice_loss_coefficient: int | float = 5.0
    segmentation_head_activation_function: str = "gelu"
    intermediate_size: int = 1024

    def __post_init__(self, **kwargs):
        self.backbone_config, kwargs = consolidate_backbone_kwargs_to_config(
            backbone_config=self.backbone_config,
            default_config_type="rf_detr_dinov2",
            default_config_kwargs={
                "num_attention_heads": 6,
                "out_features": ["stage2", "stage5", "stage8", "stage11"],
                "hidden_size": 384,
                "num_register_tokens": 0,
                "image_size": 518,
            },
            **kwargs,
        )
        super().__post_init__(**kwargs)


__all__ = ["RfDetrConfig", "RfDetrDinov2Config"]