configuration_minimax_vl_01.py

"""MiniMaxVL01 model configuration"""

from transformers.configuration_utils import PretrainedConfig
from transformers.utils import logging
from transformers.models.auto import CONFIG_MAPPING, AutoConfig
from .configuration_minimax_text_01 import MiniMaxText01Config


class MiniMaxVL01Config(PretrainedConfig):
    r"""
    This is the configuration class to store the configuration of a [`MiniMaxVL01ForConditionalGeneration`]. It is used to instantiate an
    MiniMaxVL01 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 MiniMaxVL01.

    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
    documentation from [`PretrainedConfig`] for more information.

    Args:
        vision_config (`Union[AutoConfig, dict]`,  *optional*, defaults to `CLIPVisionConfig`):
            The config object or dictionary of the vision backbone.
        text_config (`Union[AutoConfig, dict]`, *optional*, defaults to `MiniMaxText01Config`):
            The config object or dictionary of the text backbone.
        ignore_index (`int`, *optional*, defaults to -100):
            The ignore index for the loss function.
        image_token_index (`int`, *optional*, defaults to 32000):
            The image token index to encode the image prompt.
        projector_hidden_act (`str`, *optional*, defaults to `"gelu"`):
            The activation function used by the multimodal projector.
        vision_feature_select_strategy (`str`, *optional*, defaults to `"default"`):
            The feature selection strategy used to select the vision feature from the vision backbone.
            Can be one of `"default"` or `"full"`. If `"default"`, the CLS token is removed from the vision features.
            If `"full"`, the full vision features are used.
        vision_feature_layer (`int`, *optional*, defaults to -2):
            The index of the layer to select the vision feature.
        image_grid_pinpoints (`List`, *optional*, defaults to `[[336, 672], [672, 336], [672, 672], [1008, 336], [336, 1008]]`):
            A list of possible resolutions to use for processing high resolution images. Each item in the list should be a tuple or list
            of the form `(height, width)`.
        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
            Whether the model's input and output word embeddings should be tied.
        image_seq_length (`int`, *optional*, defaults to 576):
            Sequence length of one image embedding.

    Example:

    ```python
    >>> from transformers import MiniMaxVL01ForConditionalGeneration, MiniMaxVL01Config, CLIPVisionConfig, MiniMaxText01Config

    >>> # Initializing a CLIP-vision config
    >>> vision_config = CLIPVisionConfig()

    >>> # Initializing a MiniMaxText01 config
    >>> text_config = MiniMaxText01Config()

    >>> # Initializing a MiniMaxVL01 style configuration
    >>> configuration = MiniMaxVL01Config(vision_config, text_config)

    >>> # Initializing a model from the MiniMaxVL01 style configuration
    >>> model = MiniMaxVL01ForConditionalGeneration(configuration)

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

    model_type = "minimax_vl_01"

    def __init__(
        self,
        vision_config=None,
        text_config=None,
        ignore_index=-100,
        image_token_index=32000,
        projector_hidden_act="gelu",
        vision_feature_select_strategy="default",
        vision_feature_layer=-2,
        image_grid_pinpoints=None,
        tie_word_embeddings=False,
        image_seq_length=576,
        **kwargs,
    ):
        self.ignore_index = ignore_index
        self.image_token_index = image_token_index
        self.projector_hidden_act = projector_hidden_act
        self.image_seq_length = image_seq_length

        if vision_feature_select_strategy not in ["default", "full"]:
            raise ValueError(
                "vision_feature_select_strategy should be one of 'default', 'full'."
                f"Got: {vision_feature_select_strategy}"
            )

        self.vision_feature_select_strategy = vision_feature_select_strategy
        self.vision_feature_layer = vision_feature_layer
        image_grid_pinpoints = (
            image_grid_pinpoints
            if image_grid_pinpoints is not None
            else [[336, 672], [672, 336], [672, 672], [1008, 336], [336, 1008]]
        )
        self.image_grid_pinpoints = image_grid_pinpoints

        if isinstance(vision_config, dict):
            vision_config["model_type"] = (
                vision_config["model_type"] if "model_type" in vision_config else "clip_vision_model"
            )
            vision_config = CONFIG_MAPPING[vision_config["model_type"]](**vision_config)
        elif vision_config is None:
            vision_config = CONFIG_MAPPING["clip_vision_model"](
                intermediate_size=4096,
                hidden_size=1024,
                patch_size=14,
                image_size=336,
                num_hidden_layers=24,
                num_attention_heads=16,
                vocab_size=32000,
                projection_dim=768,
            )

        self.vision_config = vision_config

        if text_config is not None:
            assert "model_type" in text_config, "text_config model_type is not specified"
            text_config = MiniMaxText01Config(**text_config)
        else:
            text_config = MiniMaxText01Config()

        self.text_config = text_config

        super().__init__(tie_word_embeddings=tie_word_embeddings, **kwargs)