MMAudio-main/transformers/examples/modular-transformers/modeling_multimodal2.py

#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
#           This file was automatically generated from examples/modular-transformers/modular_multimodal2.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_multimodal2.py file directly. One of our CI enforces this.
#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨

from typing import Callable, Optional, Tuple, Union

import torch
from torch import nn

from transformers.utils import add_start_docstrings

from ...activations import ACT2FN
from ...modeling_outputs import BaseModelOutput, BaseModelOutputWithPooling
from ...modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
from ...utils import (
    add_start_docstrings_to_model_forward,
    can_return_tuple,
    logging,
    replace_return_docstrings,
    torch_int,
)
from .configuration_multimodal2 import Multimodal2Config, Multimodal2TextConfig, Multimodal2VisionConfig


logger = logging.get_logger(__name__)


def eager_attention_forward(
    module: nn.Module,
    query: torch.Tensor,
    key: torch.Tensor,
    value: torch.Tensor,
    attention_mask: Optional[torch.Tensor],
    scaling: float,
    dropout: float = 0.0,
    output_attentions: bool = True,
    **kwargs,
):
    attn_weights = torch.matmul(query, key.transpose(-1, -2)) * scaling
    if attention_mask is not None:
        attn_weights = attn_weights + attention_mask
    attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype)
    attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training)

    attn_output = torch.matmul(attn_weights, value)
    attn_output = attn_output.transpose(1, 2).contiguous()
    if not output_attentions:
        attn_weights = None
    return attn_output, attn_weights


class Multimodal2VisionAttention(nn.Module):
    """Multi-headed attention from 'Attention Is All You Need' paper"""

    def __init__(self, config: Union[Multimodal2VisionConfig, Multimodal2TextConfig]):
        super().__init__()
        self.config = config
        self.embed_dim = config.hidden_size
        self.num_heads = config.num_attention_heads
        self.head_dim = self.embed_dim // self.num_heads
        if self.head_dim * self.num_heads != self.embed_dim:
            raise ValueError(
                f"embed_dim must be divisible by num_heads (got `embed_dim`: {self.embed_dim} and `num_heads`:"
                f" {self.num_heads})."
            )
        self.scale = self.head_dim**-0.5
        self.dropout = config.attention_dropout
        self.is_causal = False

        self.k_proj = nn.Linear(self.embed_dim, self.embed_dim)
        self.v_proj = nn.Linear(self.embed_dim, self.embed_dim)
        self.q_proj = nn.Linear(self.embed_dim, self.embed_dim)
        self.out_proj = nn.Linear(self.embed_dim, self.embed_dim)

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        causal_attention_mask: Optional[torch.Tensor] = None,
        output_attentions: Optional[bool] = False,
    ) -> Tuple[torch.Tensor, Optional[torch.Tensor]]:
        """Input shape: Batch x Time x Channel"""

        batch_size, seq_length, embed_dim = hidden_states.shape

        queries = self.q_proj(hidden_states)
        keys = self.k_proj(hidden_states)
        values = self.v_proj(hidden_states)

        queries = queries.view(batch_size, seq_length, -1, self.head_dim).transpose(1, 2)
        keys = keys.view(batch_size, seq_length, -1, self.head_dim).transpose(1, 2)
        values = values.view(batch_size, seq_length, -1, self.head_dim).transpose(1, 2)
        # MULTIMODAL2_VISION text model uses both `causal_attention_mask` and `attention_mask`
        # in case FA2 kernel is called, `is_causal` should be inferred from `causal_attention_mask`
        if self.config._attn_implementation == "flash_attention_2":
            self.is_causal = causal_attention_mask is not None
        else:
            if attention_mask is not None and causal_attention_mask is not None:
                attention_mask = attention_mask + causal_attention_mask
            elif causal_attention_mask is not None:
                attention_mask = causal_attention_mask

        attention_interface: Callable = eager_attention_forward
        if self.config._attn_implementation != "eager":
            if self.config._attn_implementation == "sdpa" and output_attentions:
                logger.warning_once(
                    "`torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to "
                    'eager attention. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.'
                )
            else:
                attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]

        attn_output, attn_weights = attention_interface(
            self,
            queries,
            keys,
            values,
            attention_mask,
            is_causal=self.is_causal,
            scaling=self.scale,
            dropout=0.0 if not self.training else self.dropout,
            output_attentions=output_attentions,
        )

        attn_output = attn_output.reshape(batch_size, seq_length, embed_dim).contiguous()
        attn_output = self.out_proj(attn_output)

        if not output_attentions:
            attn_weights = None
        return attn_output, attn_weights


class Multimodal2VisionMLP(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.config = config
        self.activation_fn = ACT2FN[config.hidden_act]
        self.fc1 = nn.Linear(config.hidden_size, config.intermediate_size)
        self.fc2 = nn.Linear(config.intermediate_size, config.hidden_size)

    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
        hidden_states = self.fc1(hidden_states)
        hidden_states = self.activation_fn(hidden_states)
        hidden_states = self.fc2(hidden_states)
        return hidden_states


class Multimodal2Attention(nn.Module):
    """Multi-headed attention from 'Attention Is All You Need' paper"""

    def __init__(self, config: Union[Multimodal2VisionConfig, Multimodal2TextConfig]):
        super().__init__()
        self.config = config
        self.embed_dim = config.hidden_size
        self.num_heads = config.num_attention_heads
        self.head_dim = self.embed_dim // self.num_heads
        if self.head_dim * self.num_heads != self.embed_dim:
            raise ValueError(
                f"embed_dim must be divisible by num_heads (got `embed_dim`: {self.embed_dim} and `num_heads`:"
                f" {self.num_heads})."
            )
        self.scale = self.head_dim**-0.5
        self.dropout = config.attention_dropout
        self.is_causal = False

        self.k_proj = nn.Linear(self.embed_dim, self.embed_dim)
        self.v_proj = nn.Linear(self.embed_dim, self.embed_dim)
        self.q_proj = nn.Linear(self.embed_dim, self.embed_dim)
        self.out_proj = nn.Linear(self.embed_dim, self.embed_dim)

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        causal_attention_mask: Optional[torch.Tensor] = None,
        output_attentions: Optional[bool] = False,
    ) -> Tuple[torch.Tensor, Optional[torch.Tensor]]:
        """Input shape: Batch x Time x Channel"""

        batch_size, seq_length, embed_dim = hidden_states.shape

        queries = self.q_proj(hidden_states)
        keys = self.k_proj(hidden_states)
        values = self.v_proj(hidden_states)

        queries = queries.view(batch_size, seq_length, -1, self.head_dim).transpose(1, 2)
        keys = keys.view(batch_size, seq_length, -1, self.head_dim).transpose(1, 2)
        values = values.view(batch_size, seq_length, -1, self.head_dim).transpose(1, 2)
        # MULTIMODAL2 text model uses both `causal_attention_mask` and `attention_mask`
        # in case FA2 kernel is called, `is_causal` should be inferred from `causal_attention_mask`
        if self.config._attn_implementation == "flash_attention_2":
            self.is_causal = causal_attention_mask is not None
        else:
            if attention_mask is not None and causal_attention_mask is not None:
                attention_mask = attention_mask + causal_attention_mask
            elif causal_attention_mask is not None:
                attention_mask = causal_attention_mask

        attention_interface: Callable = eager_attention_forward
        if self.config._attn_implementation != "eager":
            if self.config._attn_implementation == "sdpa" and output_attentions:
                logger.warning_once(
                    "`torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to "
                    'eager attention. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.'
                )
            else:
                attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]

        attn_output, attn_weights = attention_interface(
            self,
            queries,
            keys,
            values,
            attention_mask,
            is_causal=self.is_causal,
            scaling=self.scale,
            dropout=0.0 if not self.training else self.dropout,
            output_attentions=output_attentions,
        )

        attn_output = attn_output.reshape(batch_size, seq_length, embed_dim).contiguous()
        attn_output = self.out_proj(attn_output)

        if not output_attentions:
            attn_weights = None
        return attn_output, attn_weights


class Multimodal2VisionEncoderLayer(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.embed_dim = config.hidden_size
        self.self_attn = Multimodal2Attention(config)
        self.layer_norm1 = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_eps)
        self.mlp = Multimodal2VisionMLP(config)
        self.layer_norm2 = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_eps)

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: torch.Tensor,
        causal_attention_mask: torch.Tensor,
        output_attentions: Optional[bool] = False,
    ) -> Tuple[torch.FloatTensor]:
        """
        Args:
            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
            attention_mask (`torch.FloatTensor`): attention mask of size
                `(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
                `(config.encoder_attention_heads,)`.
            output_attentions (`bool`, *optional*):
                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
                returned tensors for more detail.
        """
        residual = hidden_states

        hidden_states = self.layer_norm1(hidden_states)
        hidden_states, attn_weights = self.self_attn(
            hidden_states=hidden_states,
            attention_mask=attention_mask,
            causal_attention_mask=causal_attention_mask,
            output_attentions=output_attentions,
        )
        hidden_states = residual + hidden_states

        residual = hidden_states
        hidden_states = self.layer_norm2(hidden_states)
        hidden_states = self.mlp(hidden_states)
        hidden_states = residual + hidden_states

        outputs = (hidden_states,)

        if output_attentions:
            outputs += (attn_weights,)

        return outputs


class Multimodal2VisionEncoder(nn.Module):
    """
    Transformer encoder consisting of `config.num_hidden_layers` self attention layers. Each layer is a
    [`Multimodal2VisionEncoderLayer`].

    Args:
        config: Multimodal2VisionConfig
    """

    def __init__(self, config):
        super().__init__()
        self.config = config
        self.layers = nn.ModuleList([Multimodal2VisionEncoderLayer(config) for _ in range(config.num_hidden_layers)])
        self.gradient_checkpointing = False

    @can_return_tuple
    def forward(
        self,
        inputs_embeds,
        attention_mask: Optional[torch.Tensor] = None,
        causal_attention_mask: Optional[torch.Tensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
    ) -> BaseModelOutput:
        r"""
        Args:
            inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`):
                Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation.
                This is useful if you want more control over how to convert `input_ids` indices into associated vectors
                than the model's internal embedding lookup matrix.
            attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
                Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:

                - 1 for tokens that are **not masked**,
                - 0 for tokens that are **masked**.

                [What are attention masks?](../glossary#attention-mask)
            causal_attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
                Causal mask for the text model. Mask values selected in `[0, 1]`:

                - 1 for tokens that are **not masked**,
                - 0 for tokens that are **masked**.

                [What are attention masks?](../glossary#attention-mask)
            output_attentions (`bool`, *optional*):
                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
                returned tensors for more detail.
            output_hidden_states (`bool`, *optional*):
                Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors
                for more detail.
            return_dict (`bool`, *optional*):
                Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
        """
        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )

        encoder_states = () if output_hidden_states else None
        all_attentions = () if output_attentions else None

        hidden_states = inputs_embeds
        for idx, encoder_layer in enumerate(self.layers):
            if output_hidden_states:
                encoder_states = encoder_states + (hidden_states,)
            if self.gradient_checkpointing and self.training:
                layer_outputs = self._gradient_checkpointing_func(
                    encoder_layer.__call__,
                    hidden_states,
                    attention_mask,
                    causal_attention_mask,
                    output_attentions,
                )
            else:
                layer_outputs = encoder_layer(
                    hidden_states,
                    attention_mask,
                    causal_attention_mask,
                    output_attentions=output_attentions,
                )

            hidden_states = layer_outputs[0]

            if output_attentions:
                all_attentions = all_attentions + (layer_outputs[1],)

        if output_hidden_states:
            encoder_states = encoder_states + (hidden_states,)

        return BaseModelOutput(
            last_hidden_state=hidden_states,
            hidden_states=encoder_states,
            attentions=all_attentions,
        )


class Multimodal2VisionEmbeddings(nn.Module):
    def __init__(self, config: Multimodal2VisionConfig):
        super().__init__()
        self.config = config
        self.embed_dim = config.hidden_size
        self.image_size = config.image_size
        self.patch_size = config.patch_size

        self.class_embedding = nn.Parameter(torch.randn(self.embed_dim))

        self.patch_embedding = nn.Conv2d(
            in_channels=config.num_channels,
            out_channels=self.embed_dim,
            kernel_size=self.patch_size,
            stride=self.patch_size,
            bias=False,
        )

        self.num_patches = (self.image_size // self.patch_size) ** 2
        self.num_positions = self.num_patches + 1
        self.position_embedding = nn.Embedding(self.num_positions, self.embed_dim)
        self.register_buffer("position_ids", torch.arange(self.num_positions).expand((1, -1)), persistent=False)

    def interpolate_pos_encoding(self, embeddings: torch.Tensor, height: int, width: int) -> torch.Tensor:
        """
        This method allows to interpolate the pre-trained position encodings, to be able to use the model on higher resolution
        images. This method is also adapted to support torch.jit tracing.

        Adapted from:
        - https://github.com/facebookresearch/dino/blob/de9ee3df6cf39fac952ab558447af1fa1365362a/vision_transformer.py#L174-L194, and
        - https://github.com/facebookresearch/dinov2/blob/e1277af2ba9496fbadf7aec6eba56e8d882d1e35/dinov2/models/vision_transformer.py#L179-L211
        """

        num_patches = embeddings.shape[1] - 1
        position_embedding = self.position_embedding.weight.unsqueeze(0)
        num_positions = position_embedding.shape[1] - 1

        # always interpolate when tracing to ensure the exported model works for dynamic input shapes
        if not torch.jit.is_tracing() and num_patches == num_positions and height == width:
            return self.position_embedding(self.position_ids)

        class_pos_embed = position_embedding[:, :1]
        patch_pos_embed = position_embedding[:, 1:]

        dim = embeddings.shape[-1]

        new_height = height // self.patch_size
        new_width = width // self.patch_size

        sqrt_num_positions = torch_int(num_positions**0.5)
        patch_pos_embed = patch_pos_embed.reshape(1, sqrt_num_positions, sqrt_num_positions, dim)
        patch_pos_embed = patch_pos_embed.permute(0, 3, 1, 2)

        patch_pos_embed = nn.functional.interpolate(
            patch_pos_embed,
            size=(new_height, new_width),
            mode="bicubic",
            align_corners=False,
        )

        patch_pos_embed = patch_pos_embed.permute(0, 2, 3, 1).view(1, -1, dim)

        return torch.cat((class_pos_embed, patch_pos_embed), dim=1)

    def forward(self, pixel_values: torch.FloatTensor, interpolate_pos_encoding=False) -> torch.Tensor:
        batch_size, _, height, width = pixel_values.shape
        if not interpolate_pos_encoding and (height != self.image_size or width != self.image_size):
            raise ValueError(
                f"Input image size ({height}*{width}) doesn't match model ({self.image_size}*{self.image_size})."
            )
        target_dtype = self.patch_embedding.weight.dtype
        patch_embeds = self.patch_embedding(pixel_values.to(dtype=target_dtype))  # shape = [*, width, grid, grid]
        patch_embeds = patch_embeds.flatten(2).transpose(1, 2)

        class_embeds = self.class_embedding.expand(batch_size, 1, -1)
        embeddings = torch.cat([class_embeds, patch_embeds], dim=1)
        if interpolate_pos_encoding:
            embeddings = embeddings + self.interpolate_pos_encoding(embeddings, height, width)
        else:
            embeddings = embeddings + self.position_embedding(self.position_ids)
        return embeddings


MULTIMODAL2_VISION_INPUTS_DOCSTRING = r"""
    Args:
        pixel_values (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`):
            Pixel values. Padding will be ignored by default should you provide it. Pixel values can be obtained using
            [`AutoImageProcessor`]. See [`Multimodal2ImageProcessor.__call__`] for details.
        output_attentions (`bool`, *optional*):
            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
            tensors for more detail.
        output_hidden_states (`bool`, *optional*):
            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
            more detail.
        interpolate_pos_encoding (`bool`, *optional*, defaults `False`):
            Whether to interpolate the pre-trained position encodings.
        return_dict (`bool`, *optional*):
            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
"""


class Multimodal2VisionTransformer(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.config = config
        embed_dim = config.hidden_size

        self.embeddings = Multimodal2VisionEmbeddings(config)
        self.pre_layrnorm = nn.LayerNorm(embed_dim, eps=config.layer_norm_eps)
        self.encoder = Multimodal2VisionEncoder(config)
        self.post_layernorm = nn.LayerNorm(embed_dim, eps=config.layer_norm_eps)

    @can_return_tuple
    @add_start_docstrings_to_model_forward(MULTIMODAL2_VISION_INPUTS_DOCSTRING)
    @replace_return_docstrings(output_type=BaseModelOutputWithPooling, config_class=Multimodal2VisionConfig)
    def forward(
        self,
        pixel_values: Optional[torch.FloatTensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        interpolate_pos_encoding: Optional[bool] = False,
    ) -> BaseModelOutputWithPooling:
        r"""
        Returns:

        """
        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )

        if pixel_values is None:
            raise ValueError("You have to specify pixel_values")

        hidden_states = self.embeddings(pixel_values, interpolate_pos_encoding=interpolate_pos_encoding)
        hidden_states = self.pre_layrnorm(hidden_states)

        encoder_outputs: BaseModelOutput = self.encoder(
            inputs_embeds=hidden_states,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
        )

        last_hidden_state = encoder_outputs.last_hidden_state
        pooled_output = last_hidden_state[:, 0, :]
        pooled_output = self.post_layernorm(pooled_output)

        return BaseModelOutputWithPooling(
            last_hidden_state=last_hidden_state,
            pooler_output=pooled_output,
            hidden_states=encoder_outputs.hidden_states,
            attentions=encoder_outputs.attentions,
        )


class Multimodal2VisionPreTrainedModel(PreTrainedModel):
    """
    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
    models.
    """

    config_class = Multimodal2Config
    base_model_prefix = "multimodal2_vision"
    supports_gradient_checkpointing = True
    _supports_sdpa = True
    _supports_flash_attn_2 = True

    def _init_weights(self, module):
        """Initialize the weights"""
        if isinstance(module, Multimodal2VisionMLP):
            pass


MULTIMODAL2_VISION_START_DOCSTRING = "doc"


@add_start_docstrings("New doc", MULTIMODAL2_VISION_START_DOCSTRING)
class Multimodal2VisionModel(Multimodal2VisionPreTrainedModel):
    config_class = Multimodal2VisionConfig
    main_input_name = "pixel_values"
    _no_split_modules = ["Multimodal2VisionEncoderLayer"]

    def __init__(self, config: Multimodal2VisionConfig):
        super().__init__(config)
        self.vision_model = Multimodal2VisionTransformer(config)
        # Initialize weights and apply final processing
        self.post_init()

    def get_input_embeddings(self) -> nn.Module:
        return self.vision_model.embeddings.patch_embedding

    @can_return_tuple
    @add_start_docstrings_to_model_forward(MULTIMODAL2_VISION_INPUTS_DOCSTRING)
    @replace_return_docstrings(output_type=BaseModelOutputWithPooling, config_class=Multimodal2VisionConfig)
    def forward(
        self,
        pixel_values: Optional[torch.FloatTensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        interpolate_pos_encoding: bool = False,
    ) -> BaseModelOutputWithPooling:
        r"""
        Returns:

        Examples:

        ```python
        >>> from PIL import Image
        >>> import requests
        >>> from transformers import AutoProcessor, Multimodal2VisionModel

        >>> model = Multimodal2VisionModel.from_pretrained("openai/multimodal2-vit-base-patch32")
        >>> processor = AutoProcessor.from_pretrained("openai/multimodal2-vit-base-patch32")

        >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
        >>> image = Image.open(requests.get(url, stream=True).raw)

        >>> inputs = processor(images=image, return_tensors="pt")

        >>> outputs = model(**inputs)
        >>> last_hidden_state = outputs.last_hidden_state
        >>> pooled_output = outputs.pooler_output  # pooled CLS states
        ```"""

        return self.vision_model(
            pixel_values=pixel_values,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            interpolate_pos_encoding=interpolate_pos_encoding,
        )