#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
#           This file was automatically generated from src/transformers/models/cohere2_vision/modular_cohere2_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_cohere2_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 ...cache_utils import Cache
from ...generation import GenerationMixin
from ...modeling_flash_attention_utils import FlashAttentionKwargs
from ...modeling_outputs import BaseModelOutputWithPast, ModelOutput
from ...modeling_utils import PreTrainedModel
from ...processing_utils import Unpack
from ...utils import TransformersKwargs, auto_docstring
from ...utils.generic import check_model_inputs
from ..auto import AutoModel
from .configuration_cohere2_vision import Cohere2VisionConfig


class Cohere2VisionMultiModalProjector(nn.Module):
    def __init__(self, config: Cohere2VisionConfig):
        super().__init__()
        self.config = config
        self.downsample_factor = config.downsample_factor
        self.intermediate_size = config.alignment_intermediate_size
        self.linear_1 = nn.Linear(
            config.vision_config.hidden_size * (config.downsample_factor**2), self.intermediate_size, bias=True
        )
        self.act = nn.SiLU()
        self.linear_2 = nn.Linear(self.intermediate_size // 2, config.text_config.hidden_size, bias=True)

    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

    def forward(self, image_features):
        image_features = self.pixel_shuffle(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


@dataclass
@auto_docstring(
    custom_intro="""
    Base class for Cohere2Vision outputs, with hidden states and attentions.
    """
)
class Cohere2VisionModelOutputWithPast(BaseModelOutputWithPast):
    r"""
    past_key_values (`Cache`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
        Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
        `(batch_size, num_heads, sequence_length, embed_size_per_head)`)

        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


@dataclass
@auto_docstring(
    custom_intro="""
    Base class for Cohere2Vision causal language model (or autoregressive) outputs.
    """
)
class Cohere2VisionCausalLMOutputWithPast(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`):
        Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
        `(batch_size, num_heads, sequence_length, embed_size_per_head)`)

        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[list[torch.FloatTensor]] = None
    hidden_states: Optional[tuple[torch.FloatTensor]] = None
    attentions: Optional[tuple[torch.FloatTensor]] = None
    image_hidden_states: Optional[torch.FloatTensor] = None


@auto_docstring
class Cohere2VisionPreTrainedModel(PreTrainedModel):
    config: Cohere2VisionConfig
    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",
    }


@auto_docstring(
    custom_intro="""
    The Cohere2Vision model which consists of a vision backbone and a language model, without a language modeling head.
    """
)
class Cohere2VisionModel(Cohere2VisionPreTrainedModel):
    _checkpoint_conversion_mapping = {}

    def __init__(self, config: Cohere2VisionConfig):
        super().__init__(config)
        self.vision_tower = AutoModel.from_config(config.vision_config)

        self.multi_modal_projector = Cohere2VisionMultiModalProjector(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,
        image_num_patches: torch.Tensor,
    ):
        """
        Obtains image last hidden states from the vision tower and apply multimodal projection.

        Args:
            pixel_values (`torch.FloatTensor]` of shape `(batch_size, num_patches, channels, height, width)`)
               The tensors corresponding to the input images.
            image_num_patches (`torch.Tensor` of shape `(num_images)`)
                Number of patches for each image.
        Returns:
            image_features (List[`torch.Tensor`]): List of image feature tensor, each contains all the visual feature of all patches
            and are of shape `(num_patches, image_length, embed_dim)`).
        """

        image_features = self.vision_tower(pixel_values, output_hidden_states=True)
        selected_image_feature = image_features.last_hidden_state
        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 placeholdr 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: torch.LongTensor = None,
        pixel_values: torch.FloatTensor = None,
        image_num_patches: Optional[torch.Tensor] = 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,
        use_cache: Optional[bool] = None,
        cache_position: Optional[torch.LongTensor] = None,
        **kwargs: Unpack[FlashAttentionKwargs],
    ) -> Union[tuple, Cohere2VisionModelOutputWithPast]:
        r"""
        image_num_patches (`torch.Tensor` of shape `(num_images,)`):
            Number of patches per input image.
        """
        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, image_num_patches=image_num_patches)
            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 Cohere2VisionModelOutputWithPast(
            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 COHERE2_VISION model which consists of a vision backbone and a language model.
    """
)
class Cohere2VisionForConditionalGeneration(Cohere2VisionPreTrainedModel, GenerationMixin):
    _checkpoint_conversion_mapping = {}
    _tied_weights_keys = ["lm_head.weight"]

    def __init__(self, config: Cohere2VisionConfig):
        super().__init__(config)
        self.model = Cohere2VisionModel(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,
        image_num_patches: torch.Tensor,
    ):
        return self.model.get_image_features(
            pixel_values=pixel_values,
            image_num_patches=image_num_patches,
        )

    # Make modules available throught 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

    @check_model_inputs
    @auto_docstring
    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        pixel_values: Optional[torch.FloatTensor] = None,
        image_num_patches: Optional[torch.Tensor] = 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,
        labels: Optional[torch.LongTensor] = None,
        use_cache: Optional[bool] = 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, Cohere2VisionCausalLMOutputWithPast]:
        r"""
        image_num_patches (`torch.Tensor` of shape `(num_images,)`):
            Number of patches per input image.
        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, Cohere2VisionForConditionalGeneration
        >>> import torch

        >>> processor = AutoProcessor.from_pretrained("CohereLabs/command-a-vision-07-2025", use_fast=True)
        >>> model = Cohere2VisionForConditionalGeneration.from_pretrained("CohereLabs/command-a-vision-07-2025", device_map="auto")

        >>> messages = [
        ...     {
        ...         "role": "user",
        ...         "content": [
        ...             {
        ...                 "type": "image",
        ...                 "url": "https://images.pexels.com/photos/1108099/pexels-photo-1108099.jpeg",
        ...             },
        ...             {"type": "text", "text": "what is in this image?"},
        ...         ],
        ...     },
        ... ]

        >>> inputs = processor.apply_chat_template(
        ...     messages, padding=True, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt",
        ... ).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)
        ```"""
        outputs = self.model(
            input_ids=input_ids,
            pixel_values=pixel_values,
            image_num_patches=image_num_patches,
            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,
            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 Cohere2VisionCausalLMOutputWithPast(
            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__ = ["Cohere2VisionForConditionalGeneration", "Cohere2VisionPreTrainedModel", "Cohere2VisionModel"]