configuration.py

from typing import Optional

from transformers.configuration_utils import PretrainedConfig
from transformers.modeling_rope_utils import rope_config_validation

class PrismaVLVisionConfig(PretrainedConfig):
    model_type = "qwen3_vl"
    base_config_key = "vision_config"

    def __init__(
        self,
        depth=27,
        hidden_size=1152,
        hidden_act="gelu_pytorch_tanh",
        intermediate_size=4304,
        num_heads=16,
        in_channels=3,
        patch_size=16,
        spatial_merge_size=2,
        temporal_patch_size=2,
        out_hidden_size=3584,
        num_position_embeddings=2304,
        deepstack_visual_indexes=[8, 16, 24],
        initializer_range=0.02,
        **kwargs,
    ):
        super().__init__(**kwargs)

        self.depth = depth
        self.hidden_size = hidden_size
        self.hidden_act = hidden_act
        self.intermediate_size = intermediate_size
        self.num_heads = num_heads
        self.in_channels = in_channels
        self.patch_size = patch_size
        self.spatial_merge_size = spatial_merge_size
        self.temporal_patch_size = temporal_patch_size
        self.out_hidden_size = out_hidden_size
        self.num_position_embeddings = num_position_embeddings
        self.initializer_range = initializer_range
        self.deepstack_visual_indexes = deepstack_visual_indexes


class PrismaVLTextConfig(PretrainedConfig):
    r"""
    This is the configuration class to store the configuration of a [`PrismaVLTextModel`]. It is used to instantiate a
    Prisma-VL model according to the specified arguments, defining the model architecture. Instantiating a configuration
    with the defaults will yield a similar configuration to that of
    Prisma-VL-4B-Instruct [Qwen/Prisma-VL-4B-Instruct](https://huggingface.co/Qwen/Prisma-VL-4B-Instruct).

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

    Args:
        vocab_size (`int`, *optional*, defaults to 151936):
            Vocabulary size of the PrismaVL model. Defines the number of different tokens that can be represented by the
            `inputs_ids` passed when calling [`PrismaVLModel`]
        hidden_size (`int`, *optional*, defaults to 4096):
            Dimension of the hidden representations.
        intermediate_size (`int`, *optional*, defaults to 22016):
            Dimension of the MLP representations.
        num_hidden_layers (`int`, *optional*, defaults to 32):
            Number of hidden layers in the Transformer encoder.
        num_attention_heads (`int`, *optional*, defaults to 32):
            Number of attention heads for each attention layer in the Transformer encoder.
        num_key_value_heads (`int`, *optional*, defaults to 32):
            This is the number of key_value heads that should be used to implement Grouped Query Attention. If
            `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
            `num_key_value_heads=1` the model will use Multi Query Attention (MQA) otherwise GQA is used. When
            converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
            by meanpooling all the original heads within that group. For more details, check out [this
            paper](https://huggingface.co/papers/2305.13245). If it is not specified, will default to `32`.
        head_dim (`int`, *optional*, defaults to 128):
            The dimension of the head. If not specified, will default to `hidden_size // num_attention_heads`.
        hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
            The non-linear activation function (function or string) in the decoder.
        max_position_embeddings (`int`, *optional*, defaults to 128000):
            The maximum sequence length that this model might ever be used with.
        initializer_range (`float`, *optional*, defaults to 0.02):
            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
        rms_norm_eps (`float`, *optional*, defaults to 1e-06):
            The epsilon used by the rms normalization layers.
        use_cache (`bool`, *optional*, defaults to `True`):
            Whether or not the model should return the last key/values attentions (not used by all models). Only
            relevant if `config.is_decoder=True`.
        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
            Whether the model's input and output word embeddings should be tied.
        rope_theta (`float`, *optional*, defaults to 5000000.0):
            The base period of the RoPE embeddings.
        rope_scaling (`Dict`, *optional*):
            Dictionary containing the scaling configuration for the RoPE embeddings. Contains parameters for
            scaling RoPE to work with longer sequences.
        attention_bias (`bool`, defaults to `False`, *optional*, defaults to `False`):
            Whether to use a bias in the query, key, value and output projection layers during self-attention.
        attention_dropout (`float`, *optional*, defaults to 0.0):
            The dropout ratio for the attention probabilities.

    ```python
    >>> from transformers import PrismaVLTextModel, PrismaVLTextConfig

    >>> # Initializing a PrismaVL style configuration
    >>> configuration = PrismaVLTextConfig()

    >>> # Initializing a model from the Prisma-VL-7B style configuration
    >>> model = PrismaVLTextModel(configuration)

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

    model_type = "qwen3_vl_text"
    base_config_key = "text_config"

    def __init__(
        self,
        vocab_size: Optional[int] = 151936,
        hidden_size: Optional[int] = 4096,
        intermediate_size: Optional[int] = 22016,
        num_hidden_layers: Optional[int] = 32,
        num_attention_heads: Optional[int] = 32,
        num_key_value_heads: Optional[int] = 32,
        head_dim: Optional[int] = 128,
        hidden_act: Optional[str] = "silu",
        max_position_embeddings: Optional[int] = 128000,
        initializer_range: Optional[float] = 0.02,
        rms_norm_eps: Optional[float] = 1e-6,
        use_cache: Optional[bool] = True,
        tie_word_embeddings: Optional[bool] = False,
        rope_theta: Optional[float] = 5000000.0,
        rope_scaling: Optional[dict] = None,
        attention_bias: Optional[bool] = False,
        attention_dropout: Optional[float] = 0.0,
        **kwargs,
    ):
        self.vocab_size = vocab_size
        self.max_position_embeddings = max_position_embeddings
        self.hidden_size = hidden_size
        self.intermediate_size = intermediate_size
        self.num_hidden_layers = num_hidden_layers
        self.num_attention_heads = num_attention_heads

        # for backward compatibility
        if num_key_value_heads is None:
            num_key_value_heads = num_attention_heads

        self.num_key_value_heads = num_key_value_heads
        self.head_dim = head_dim
        self.hidden_act = hidden_act
        self.initializer_range = initializer_range
        self.rms_norm_eps = rms_norm_eps
        self.use_cache = use_cache
        self.attention_bias = attention_bias
        self.attention_dropout = attention_dropout
        self.rope_theta = rope_theta
        self.rope_scaling = rope_scaling

        # Validate the correctness of rotary position embeddings parameters
        rope_config_validation(self, ignore_keys={"mrope_section", "mrope_interleaved"})

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


class PrismaVLConfig(PretrainedConfig):
    r"""
    This is the configuration class to store the configuration of a [`PrismaVLModel`]. It is used to instantiate a
    Prisma-VL model according to the specified arguments, defining the model architecture. Instantiating a configuration
    with the defaults will yield a similar configuration to that of
    Prisma-VL-4B-Instruct [Qwen/Prisma-VL-4B-Instruct](https://huggingface.co/Qwen/Prisma-VL-4B-Instruct).

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


    Args:
        text_config (`Union[PreTrainedConfig, dict]`, *optional*, defaults to `PrismaVLTextConfig`):
            The config object or dictionary of the text backbone.
        vision_config (`Union[PreTrainedConfig, dict]`,  *optional*, defaults to `PrismaVLVisionConfig`):
            The config object or dictionary of the vision backbone.
        image_token_id (`int`, *optional*, defaults to 151655):
            The image token index to encode the image prompt.
        video_token_id (`int`, *optional*, defaults to 151656):
            The video token index to encode the image prompt.
        vision_start_token_id (`int`, *optional*, defaults to 151652):
            The start token index to encode the image prompt.
        vision_end_token_id (`int`, *optional*, defaults to 151653):
            The end token index to encode the image prompt.
        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
            Whether to tie the word embeddings.

    ```python
    >>> from transformers import PrismaVLForConditionalGeneration, PrismaVLConfig

    >>> # Initializing a Prisma-VL style configuration
    >>> configuration = PrismaVLConfig()

    >>> # Initializing a model from the Prisma-VL-4B style configuration
    >>> model = PrismaVLForConditionalGeneration(configuration)

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

    model_type = "qwen3_vl"
    sub_configs = {"vision_config": PrismaVLVisionConfig, "text_config": PrismaVLTextConfig}
    keys_to_ignore_at_inference = ["past_key_values"]

    def __init__(
        self,
        text_config=None,
        vision_config=None,
        image_token_id=151655,
        video_token_id=151656,
        vision_start_token_id=151652,
        vision_end_token_id=151653,
        tie_word_embeddings=False,
        **kwargs,
    ):
        if isinstance(vision_config, dict):
            self.vision_config = self.sub_configs["vision_config"](**vision_config)
        elif vision_config is None:
            self.vision_config = self.sub_configs["vision_config"]()

        if isinstance(text_config, dict):
            self.text_config = self.sub_configs["text_config"](**text_config)
        elif text_config is None:
            self.text_config = self.sub_configs["text_config"]()

        self.image_token_id = image_token_id
        self.video_token_id = video_token_id
        self.vision_start_token_id = vision_start_token_id
        self.vision_end_token_id = vision_end_token_id
        super().__init__(**kwargs, tie_word_embeddings=tie_word_embeddings)


__all__ = ["PrismaVLConfig", "PrismaVLTextConfig"]