Remove AriaVisionModel by just using Idefics3

src/transformers/__init__.py

@@ -172,7 +172,6 @@
     "models.aria": [
         "AriaConfig",
         "AriaTextConfig",
-        "AriaVisionConfig",
     ],
     "models.audio_spectrogram_transformer": [
         "ASTConfig",
@@ -2461,6 +2460,8 @@
             "Idefics3Model",
             "Idefics3PreTrainedModel",
             "Idefics3Processor",
+            "Idefics3VisionTransformer",
+            "Idefics3VisionConfig",
         ]
     )
     _import_structure["models.imagegpt"].extend(
@@ -5022,7 +5023,6 @@
     from .models.aria import (
         AriaConfig,
         AriaTextConfig,
-        AriaVisionConfig,
     )
     from .models.audio_spectrogram_transformer import (
         ASTConfig,
@@ -7178,6 +7178,8 @@
             Idefics3Model,
             Idefics3PreTrainedModel,
             Idefics3Processor,
+            Idefics3VisionTransformer,
+            Idefics3VisionConfig,
         )
         from .models.imagegpt import (
             ImageGPTForCausalImageModeling,

src/transformers/models/aria/__init__.py

@@ -17,7 +17,7 @@
 
 
 _import_structure = {
-    "configuration_aria": ["AriaConfig", "AriaForCausalLM", "AriaTextConfig", "AriaVisionConfig"],
+    "configuration_aria": ["AriaConfig", "AriaForCausalLM", "AriaTextConfig"],
     "modeling_aria": ["AriaForConditionalGeneration", "AriaPreTrainedModel"],
     "processing_aria": ["AriaProcessor"],
 }
@@ -41,13 +41,12 @@
     ]
     _import_structure["configuration_aria"] = [
         "AriaConfig",
-        "AriaVisionConfig",
         "AriaTextConfig",
     ]
 
 
 if TYPE_CHECKING:
-    from .configuration_aria import AriaConfig, AriaTextConfig, AriaVisionConfig
+    from .configuration_aria import AriaConfig, AriaTextConfig
 
     try:
         if not is_torch_available():

src/transformers/models/aria/configuration_aria.py

@@ -4,113 +4,11 @@
 #             the file from the modular. If any change should be done, please apply the change to the
 #                          modular_aria.py file directly. One of our CI enforces this.
 #                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
-import os
-from typing import Union
+
 
 from ...configuration_utils import PretrainedConfig
 from ...modeling_rope_utils import rope_config_validation
-from ...utils import logging
-
-
-logger = logging.get_logger(__name__)
-
-
-class AriaVisionConfig(PretrainedConfig):
-    r"""
-        This is the configuration class to store the configuration of a [`AriaVisionModel`]. It is used to instantiate a
-        Aria vision encoder according to the specified arguments, defining the model architecture. Instantiating a
-        configuration with the defaults will yield a similar configuration to that of the vision encoder of the Aria
-        [google/aria-base-patch16-224](https://huggingface.co/google/aria-base-patch16-224) architecture.
-
-        Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
-        documentation from [`PretrainedConfig`] for more information.
-
-        Args:
-            hidden_size (`int`, *optional*, defaults to 768):
-                Dimensionality of the encoder layers and the pooler layer.
-            intermediate_size (`int`, *optional*, defaults to 3072):
-                Dimensionality of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.
-            num_hidden_layers (`int`, *optional*, defaults to 12):
-                Number of hidden layers in the Transformer encoder.
-            num_attention_heads (`int`, *optional*, defaults to 12):
-                Number of attention heads for each attention layer in the Transformer encoder.
-            num_channels (`int`, *optional*, defaults to 3):
-                Number of channels in the input images.
-            image_size (`int`, *optional*, defaults to 224):
-                The size (resolution) of each image.
-            patch_size (`int`, *optional*, defaults to 16):
-                The size (resolution) of each patch.
-            hidden_act (`str` or `function`, *optional*, defaults to `"gelu_pytorch_tanh"`):
-                The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`,
-                `"relu"`, `"selu"` and `"gelu_new"` `"quick_gelu"` are supported.
-            layer_norm_eps (`float`, *optional*, defaults to 1e-06):
-                The epsilon used by the layer normalization layers.
-            attention_dropout (`float`, *optional*, defaults to 0.0):
-                The dropout ratio for the attention probabilities.
-
-        Example:
-
-        ```python
-        >>> from transformers import AriaVisionConfig, AriaVisionModel
-
-        >>> # Initializing a AriaVisionConfig with google/aria-base-patch16-224 style configuration
-        >>> configuration = AriaVisionConfig()
-
-        >>> # Initializing a AriaVisionModel (with random weights) from the google/aria-base-patch16-224 style configuration
-        >>> model = AriaVisionModel(configuration)
-
-        >>> # Accessing the model configuration
-        >>> configuration = model.config
-        ```
-    Configuration class for AriaVisionModel."""
-
-    model_type = "aria_vision_model"
-
-    def __init__(
-        self,
-        hidden_size=768,
-        intermediate_size=3072,
-        num_hidden_layers=12,
-        num_attention_heads=12,
-        num_channels=3,
-        image_size=224,
-        patch_size=16,
-        hidden_act="gelu_pytorch_tanh",
-        layer_norm_eps=1e-6,
-        attention_dropout=0.0,
-        **kwargs,
-    ):
-        super().__init__(**kwargs)
-
-        self.hidden_size = hidden_size
-        self.intermediate_size = intermediate_size
-        self.num_hidden_layers = num_hidden_layers
-        self.num_attention_heads = num_attention_heads
-        self.num_channels = num_channels
-        self.patch_size = patch_size
-        self.image_size = image_size
-        self.attention_dropout = attention_dropout
-        self.layer_norm_eps = layer_norm_eps
-        self._supports_sdpa = False
-        self.hidden_act = hidden_act
-
-    @classmethod
-    def from_pretrained(cls, pretrained_model_name_or_path: Union[str, os.PathLike], **kwargs) -> "PretrainedConfig":
-        cls._set_token_in_kwargs(kwargs)
-
-        config_dict, kwargs = cls.get_config_dict(pretrained_model_name_or_path, **kwargs)
-
-        # get the vision config dict if we are loading from AriaConfig
-        if config_dict.get("model_type") == "aria":
-            config_dict = config_dict["vision_config"]
-
-        if "model_type" in config_dict and hasattr(cls, "model_type") and config_dict["model_type"] != cls.model_type:
-            logger.warning(
-                f"You are using a model of type {config_dict['model_type']} to instantiate a model of type "
-                f"{cls.model_type}. This is not supported for all configurations of models and can yield errors."
-            )
-
-        return cls.from_dict(config_dict, **kwargs)
+from ..auto import CONFIG_MAPPING
 
 
 class AriaTextConfig(PretrainedConfig):
@@ -245,7 +143,6 @@ def __init__(
         image_token_index=32000,
         **kwargs,
     ):
-        super().__init__(**kwargs)
         self.ignore_index = ignore_index
         self.image_token_index = image_token_index
 
@@ -257,17 +154,20 @@ def __init__(
                 4900: 256,
             }
         self.projector_patch_to_query_dict = {int(k): int(v) for k, v in projector_patch_to_query_dict.items()}
-        if vision_config is None:
-            vision_config = AriaVisionConfig()
-        if text_config is None:
-            text_config = AriaTextConfig()
 
-        if isinstance(vision_config, dict) and "model_type" in vision_config:
-            vision_config = AriaVisionConfig(**vision_config)
+        if isinstance(vision_config, dict):
+            vision_config["model_type"] = "idefics3_vision"
+            vision_config = CONFIG_MAPPING[vision_config["model_type"]](**vision_config)
+        elif vision_config is None:
+            vision_config = CONFIG_MAPPING["idefics3_vision"]()
 
         self.vision_config = vision_config
 
         if isinstance(text_config, dict) and "model_type" in text_config:
             text_config = AriaTextConfig(**text_config)
+        elif text_config is None:
+            text_config = AriaTextConfig()
 
         self.text_config = text_config
+
+        super().__init__(**kwargs)