#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
#           This file was automatically generated from src/transformers/models/aya_vision/modular_aya_vision.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_aya_vision.py file directly. One of our CI enforces this.
#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
# coding=utf-8
# Copyright 2025 the Cohere 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 dataclasses import dataclass
from typing import Optional, Union

import torch
from torch import nn

from ...activations import ACT2FN
from ...cache_utils import Cache
from ...generation import GenerationMixin
from ...modeling_outputs import BaseModelOutputWithPast, ModelOutput
from ...modeling_utils import PreTrainedModel
from ...processing_utils import Unpack
from ...utils import TransformersKwargs, auto_docstring, can_return_tuple
from ...utils.generic import check_model_inputs
from ..auto import AutoModel
from .configuration_aya_vision import AyaVisionConfig


class AyaVisionMultiModalProjector(nn.Module):
    def __init__(self, config: AyaVisionConfig):
        super().__init__()
        self.config = config
        self.downsample_factor = config.downsample_factor
        self.alignment_intermediate_size = getattr(
            config, "alignment_intermediate_size", config.text_config.hidden_size
        )
        self.layernorm = nn.LayerNorm(
            config.vision_config.hidden_size * (config.downsample_factor**2), eps=config.adapter_layer_norm_eps
        )

        self.linear_1 = nn.Linear(
            config.vision_config.hidden_size * (config.downsample_factor**2),
            self.alignment_intermediate_size,
            bias=True,
        )

        self.act = ACT2FN["silu"]  # SwiGLU uses SiLU activation
        # For SwiGLU, project down to half size since we split intermediate dim
        self.linear_2 = nn.Linear(self.alignment_intermediate_size // 2, config.text_config.hidden_size, bias=True)

    def forward(self, image_features):
        image_features = self.pixel_shuffle(image_features)
        image_features = self.layernorm(image_features)
        hidden_states = self.linear_1(image_features)

        # Split along last dimension and apply SwiGLU
        x, gate = hidden_states.chunk(2, dim=-1)
        hidden_states = self.act(gate) * x

        hidden_states = self.linear_2(hidden_states)
        return hidden_states

    def pixel_shuffle(self, image_features):  # B, S, D
        batch_size, seq_length, feature_dim = image_features.shape
        height = width = int(seq_length**0.5)
        image_features = image_features.reshape(image_features.shape[0], width, height, -1)
        channels = image_features.shape[-1]
        image_features = image_features.reshape(
            batch_size, width, int(height / self.downsample_factor), int(channels * self.downsample_factor)
        )
        image_features = image_features.permute(0, 2, 1, 3)
        image_features = image_features.reshape(
            batch_size, int(height / self.downsample_factor), int(width / self.downsample_factor), -1
        )
        image_features = image_features.permute(0, 2, 1, 3)
        return image_features


@auto_docstring
class AyaVisionPreTrainedModel(PreTrainedModel):
    config: AyaVisionConfig
    base_model_prefix = ""
    supports_gradient_checkpointing = True
    _skip_keys_device_placement = "past_key_values"

    _supports_flash_attn = True
    _supports_sdpa = True
    _can_compile_fullgraph = False
    _supports_flex_attn = True
    _supports_attention_backend = True
    _can_record_outputs = {
        "hidden_states": "DecoderLayer",
        "attentions": "Attention",
    }


@dataclass
@auto_docstring(
    custom_intro="""
    Base class for AyaVision causal language model (or autoregressive) outputs.
    """
)
class AyaVisionCausalLMOutputWithPast(ModelOutput):
    r"""
    loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
        Language modeling loss (for next-token prediction).
    logits (`torch.FloatTensor` of shape `(batch_size, sequence_length, config.vocab_size)`):
        Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
    past_key_values (`Cache`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
        It is a [`~cache_utils.Cache`] instance. For more details, see our [kv cache guide](https://huggingface.co/docs/transformers/en/kv_cache).

        Contains pre-computed hidden-states (key and values in the self-attention blocks) that can be used (see
        `past_key_values` input) to speed up sequential decoding.
    image_hidden_states (`torch.FloatTensor`, *optional*):
        A `torch.FloatTensor` of size `(batch_size, num_images, sequence_length, hidden_size)`.
        image_hidden_states of the model produced by the vision encoder and after projecting the last hidden state.
    """

    loss: Optional[torch.FloatTensor] = None
    logits: Optional[torch.FloatTensor] = None
    past_key_values: Optional[Cache] = None
    hidden_states: Optional[tuple[torch.FloatTensor]] = None
    attentions: Optional[tuple[torch.FloatTensor]] = None
    image_hidden_states: Optional[torch.FloatTensor] = None


@dataclass
@auto_docstring(
    custom_intro="""
    Base class for AyaVision outputs, with hidden states and attentions.
    """
)
class AyaVisionModelOutputWithPast(BaseModelOutputWithPast):
    r"""
    past_key_values (`Cache`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
        It is a [`~cache_utils.Cache`] instance. For more details, see our [kv cache guide](https://huggingface.co/docs/transformers/en/kv_cache).

        Contains pre-computed hidden-states (key and values in the self-attention blocks) that can be used (see
        `past_key_values` input) to speed up sequential decoding.
    image_hidden_states (`torch.FloatTensor`, *optional*):
        A `torch.FloatTensor` of size `(batch_size, num_images, sequence_length, hidden_size)`.
        image_hidden_states of the model produced by the vision encoder and after projecting the last hidden state.
    """

    image_hidden_states: Optional[torch.FloatTensor] = None


@auto_docstring(
    custom_intro="""
    The AyaVision model which consists of a vision backbone and a language model, without a language modeling head.
    """
)
class AyaVisionModel(AyaVisionPreTrainedModel):
    _checkpoint_conversion_mapping = {"language_model.model": "language_model"}

    def __init__(self, config: AyaVisionConfig):
        super().__init__(config)
        self.vision_tower = AutoModel.from_config(config.vision_config)

        self.multi_modal_projector = AyaVisionMultiModalProjector(config)
        self.language_model = AutoModel.from_config(config.text_config)
        self.post_init()

    def get_input_embeddings(self):
        return self.language_model.get_input_embeddings()

    def set_input_embeddings(self, value):
        self.language_model.set_input_embeddings(value)

    def set_decoder(self, decoder):
        self.language_model = decoder

    def get_decoder(self):
        return self.language_model

    def get_image_features(
        self,
        pixel_values: torch.FloatTensor,
        vision_feature_layer: Optional[Union[int, list[int]]] = None,
        vision_feature_select_strategy: Optional[str] = None,
        **kwargs,
    ):
        """
        Obtains image last hidden states from the vision tower and apply multimodal projection.

        Args:
            pixel_values (`torch.FloatTensor]` of shape `(batch_size, channels, height, width)`):
               The tensors corresponding to the input images.
            vision_feature_layer (`Union[int, list[int]]`, *optional*):
                The index of the layer to select the vision feature. If multiple indices are provided,
                the vision feature of the corresponding indices will be concatenated to form the
                vision features.
            vision_feature_select_strategy (`str`, *optional*):
                The feature selection strategy used to select the vision feature from the vision backbone.
                Can be one of `"default"` or `"full"`
        Returns:
            image_features (`torch.Tensor`): Image feature tensor of shape `(num_images, image_length, embed_dim)`).
        """
        vision_feature_layer = (
            vision_feature_layer if vision_feature_layer is not None else self.config.vision_feature_layer
        )
        vision_feature_select_strategy = (
            vision_feature_select_strategy
            if vision_feature_select_strategy is not None
            else self.config.vision_feature_select_strategy
        )

        if vision_feature_select_strategy not in ["default", "full"]:
            raise ValueError(f"Unexpected select feature strategy: {self.config.vision_feature_select_strategy}")

        kwargs = {k: v for k, v in kwargs.items() if v is not None}
        # this is not memory efficient at all (output_hidden_states=True) will save all the hidden states.
        image_outputs = self.vision_tower(pixel_values, output_hidden_states=True, **kwargs)

        # If we have one vision feature layer, return the corresponding hidden states,
        # otherwise, select the hidden states of each feature layer and concatenate them
        if isinstance(vision_feature_layer, int):
            selected_image_feature = image_outputs.hidden_states[vision_feature_layer]
            if vision_feature_select_strategy == "default":
                selected_image_feature = selected_image_feature[:, 1:]
        else:
            hs_pool = [image_outputs.hidden_states[layer_idx] for layer_idx in vision_feature_layer]
            # For default; crop CLS from each hidden state in the hidden state pool
            if vision_feature_select_strategy == "default":
                hs_pool = [hs[:, 1:] for hs in hs_pool]
            selected_image_feature = torch.cat(hs_pool, dim=-1)

        image_features = self.multi_modal_projector(selected_image_feature)
        return image_features

    def get_placeholder_mask(
        self, input_ids: torch.LongTensor, inputs_embeds: torch.FloatTensor, image_features: torch.FloatTensor
    ):
        """
        Obtains multimodal placeholder mask from `input_ids` or `inputs_embeds`, and checks that the placeholder token count is
        equal to the length of multimodal features. If the lengths are different, an error is raised.
        """
        if input_ids is None:
            special_image_mask = inputs_embeds == self.get_input_embeddings()(
                torch.tensor(self.config.image_token_id, dtype=torch.long, device=inputs_embeds.device)
            )
            special_image_mask = special_image_mask.all(-1)
        else:
            special_image_mask = input_ids == self.config.image_token_id

        n_image_tokens = special_image_mask.sum()
        special_image_mask = special_image_mask.unsqueeze(-1).expand_as(inputs_embeds).to(inputs_embeds.device)
        n_image_features = image_features.shape[0] * image_features.shape[1]
        if inputs_embeds[special_image_mask].numel() != image_features.numel():
            raise ValueError(
                f"Image features and image tokens do not match: tokens: {n_image_tokens}, features {n_image_features}"
            )
        return special_image_mask

    @check_model_inputs
    @auto_docstring
    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        pixel_values: Optional[torch.FloatTensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[Cache] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        vision_feature_layer: Optional[Union[int, list[int]]] = None,
        vision_feature_select_strategy: Optional[str] = None,
        use_cache: Optional[bool] = None,
        cache_position: Optional[torch.LongTensor] = None,
        **kwargs: Unpack[TransformersKwargs],
    ) -> Union[tuple, AyaVisionModelOutputWithPast]:
        vision_feature_layer = (
            vision_feature_layer if vision_feature_layer is not None else self.config.vision_feature_layer
        )
        vision_feature_select_strategy = (
            vision_feature_select_strategy
            if vision_feature_select_strategy is not None
            else self.config.vision_feature_select_strategy
        )

        if (input_ids is None) ^ (inputs_embeds is not None):
            raise ValueError("You must specify exactly one of input_ids or inputs_embeds")

        if inputs_embeds is None:
            inputs_embeds = self.get_input_embeddings()(input_ids)

        if pixel_values is not None:
            image_features = self.get_image_features(
                pixel_values=pixel_values,
                vision_feature_layer=vision_feature_layer,
                vision_feature_select_strategy=vision_feature_select_strategy,
            )
            image_features = image_features.to(inputs_embeds.device, inputs_embeds.dtype)
            special_image_mask = self.get_placeholder_mask(
                input_ids, inputs_embeds=inputs_embeds, image_features=image_features
            )
            inputs_embeds = inputs_embeds.masked_scatter(special_image_mask, image_features)

        outputs = self.language_model(
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_values=past_key_values,
            inputs_embeds=inputs_embeds,
            use_cache=use_cache,
            cache_position=cache_position,
            **kwargs,
        )

        return AyaVisionModelOutputWithPast(
            last_hidden_state=outputs.last_hidden_state,
            past_key_values=outputs.past_key_values,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
            image_hidden_states=image_features if pixel_values is not None else None,
        )


@auto_docstring(
    custom_intro="""
    The AYA_VISION model which consists of a vision backbone and a language model.
    """
)
class AyaVisionForConditionalGeneration(AyaVisionPreTrainedModel, GenerationMixin):
    _checkpoint_conversion_mapping = {
        "^language_model.model": "model.language_model",
        "^vision_tower": "model.vision_tower",
        "^multi_modal_projector": "model.multi_modal_projector",
        "^language_model.lm_head": "lm_head",
    }
    _tied_weights_keys = ["lm_head.weight"]

    def __init__(self, config: AyaVisionConfig):
        super().__init__(config)
        self.model = AyaVisionModel(config)
        self.lm_head = nn.Linear(config.text_config.hidden_size, config.text_config.vocab_size, bias=False)
        self.post_init()

    def get_input_embeddings(self):
        return self.model.get_input_embeddings()

    def set_input_embeddings(self, value):
        self.model.set_input_embeddings(value)

    def get_output_embeddings(self) -> nn.Module:
        return self.lm_head

    def set_decoder(self, decoder):
        self.model.set_decoder(decoder)

    def get_decoder(self):
        return self.model.get_decoder()

    def get_image_features(
        self,
        pixel_values: torch.FloatTensor,
        vision_feature_layer: Optional[Union[int, list[int]]] = None,
        vision_feature_select_strategy: Optional[str] = None,
        **kwargs,
    ):
        return self.model.get_image_features(
            pixel_values=pixel_values,
            vision_feature_layer=vision_feature_layer,
            vision_feature_select_strategy=vision_feature_select_strategy,
            **kwargs,
        )

    # Make modules available through conditional class for BC
    @property
    def language_model(self):
        return self.model.language_model

    @property
    def vision_tower(self):
        return self.model.vision_tower

    @property
    def multi_modal_projector(self):
        return self.model.multi_modal_projector

    @can_return_tuple
    @auto_docstring
    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        pixel_values: Optional[torch.FloatTensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[Cache] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        vision_feature_layer: Optional[Union[int, list[int]]] = None,
        vision_feature_select_strategy: Optional[str] = None,
        labels: Optional[torch.LongTensor] = None,
        cache_position: Optional[torch.LongTensor] = None,
        logits_to_keep: Union[int, torch.Tensor] = 0,
        image_sizes: Optional[torch.Tensor] = None,
        **kwargs: Unpack[TransformersKwargs],
    ) -> Union[tuple, AyaVisionCausalLMOutputWithPast]:
        r"""
        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
            config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.

        Example:

        ```python
        >>> from transformers import AutoProcessor, AyaVisionForConditionalGeneration
        >>> import torch

        >>> torch_device = "cuda:0"
        >>> processor = AutoProcessor.from_pretrained("CohereForAI/aya-vision-8b", use_fast=True)
        >>> model = AyaVisionForConditionalGeneration.from_pretrained("CohereForAI/aya-vision-8b", device_map=torch_device)

        >>> messages = [
        ...     {
        ...         "role": "user",
        ...         "content": [
        ...             {
        ...                 "type": "image",
        ...                 "url": "https://pbs.twimg.com/media/Fx7YvfQWYAIp6rZ?format=jpg&name=medium",
        ...             },
        ...             {"type": "text", "text": "चित्र में लिखा पाठ क्या कहता है?"},
        ...         ],
        ...     }
        ... ]

        >>> inputs = processor.apply_chat_template(
        ...     messages, padding=True, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", device=torch_device
        ... ).to(model.device)

        >>> gen_tokens = model.generate(**inputs, max_new_tokens=300, do_sample=True, temperature=0.3)
        >>> processor.tokenizer.decode(gen_tokens[0][inputs.input_ids.shape[1]:], skip_special_tokens=True)
        ```"""
        vision_feature_layer = (
            vision_feature_layer if vision_feature_layer is not None else self.config.vision_feature_layer
        )
        vision_feature_select_strategy = (
            vision_feature_select_strategy
            if vision_feature_select_strategy is not None
            else self.config.vision_feature_select_strategy
        )

        outputs = self.model(
            input_ids=input_ids,
            pixel_values=pixel_values,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_values=past_key_values,
            inputs_embeds=inputs_embeds,
            vision_feature_layer=vision_feature_layer,
            vision_feature_select_strategy=vision_feature_select_strategy,
            cache_position=cache_position,
            image_sizes=image_sizes,
            **kwargs,
        )

        hidden_states = outputs[0]
        # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
        slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep
        logits = self.lm_head(hidden_states[:, slice_indices, :])

        loss = None
        if labels is not None:
            loss = self.loss_function(
                logits=logits, labels=labels, vocab_size=self.config.text_config.vocab_size, **kwargs
            )

        return AyaVisionCausalLMOutputWithPast(
            loss=loss,
            logits=logits,
            past_key_values=outputs.past_key_values,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
            image_hidden_states=outputs.image_hidden_states,
        )

    def prepare_inputs_for_generation(
        self,
        input_ids,
        past_key_values=None,
        inputs_embeds=None,
        pixel_values=None,
        attention_mask=None,
        cache_position=None,
        logits_to_keep=None,
        **kwargs,
    ):
        # Overwritten -- in specific circumstances we don't want to forward image inputs to the model

        model_inputs = super().prepare_inputs_for_generation(
            input_ids,
            past_key_values=past_key_values,
            inputs_embeds=inputs_embeds,
            attention_mask=attention_mask,
            cache_position=cache_position,
            logits_to_keep=logits_to_keep,
            **kwargs,
        )

        if cache_position[0] == 0:
            # If we're in cached decoding stage, pixel values should be None because input ids do not contain special image token anymore
            # Otherwise we need pixel values to be passed to model
            model_inputs["pixel_values"] = pixel_values

        return model_inputs


__all__ = ["AyaVisionForConditionalGeneration", "AyaVisionPreTrainedModel", "AyaVisionModel"]