多模态图像检索BLIP模型（环境搭建、数据下载与模型测试）

1、 环境搭建

conda create -n blip python=3.8 -y
conda activate blip

pip install torch==1.9.1+cu111 torchvision==0.10.1+cu111 -f https://download.pytorch.org/whl/torch_stable.html

pip install torch==1.9.1+cu102 torchvision==0.10.1+cu102  torchaudio===0.9.1 -f https://download.pytorch.org/whl/torch_stable.html -i http://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com

# CUDA 10.2
conda install pytorch==1.9.1 torchvision==0.10.1 torchaudio==0.9.1 cudatoolkit=10.2 -c pytorch

pip3 install torch==1.8.2+cu102 torchvision==0.9.2+cu102 torchaudio===0.8.2 -f https://download.pytorch.org/whl/lts/1.8/torch_lts.html  -i http://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com

# CUDA 117
pip install torch==1.13.0+cu117 torchvision==0.14.0+cu117 torchaudio==0.13.0 --extra-index-url https://download.pytorch.org/whl/cu117  -i http://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com

import torch

if torch.cuda.is_available():
    print("GPU 可用")
else:
    print("GPU 不可用")

cd /media/lhy/ImageRetrieval/BLIP-main
pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple
pip install ruamel.yaml==0.15.87  -i https://pypi.tuna.tsinghua.edu.cn/simple
pip install opencv-python   -i https://pypi.tuna.tsinghua.edu.cn/simple

2、 测试

1、下载预训练模型model_base_retrieval_coco.pth

2、下载语料库bert-base-uncased

3、下载2014coco数据val2014.zip

4、Image-Text Retrieval

Download COCO and Flickr30k datasets from the original websites, and set ‘image_root’ in configs/retrieval_{dataset}.yaml accordingly.
To evaluate the finetuned BLIP model on COCO, run:

python -m torch.distributed.run --nproc_per_node=8 train_retrieval.py \
--config ./configs/retrieval_coco.yaml \
--output_dir output/retrieval_coco \
--evaluate

我们的环境测试：

python -m torch.distributed.run --nproc_per_node=2 train_retrieval.py \
--config ./configs/retrieval_coco.yaml \
--output_dir output/retrieval_coco \
--evaluate

要使用 8 个 A100 GPU 微调预训练检查点，请首先在 configs/retrieval_coco.yaml 中将“预训练”设置为“ https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base.pth ”。然后运行：

python -m torch.distributed.run --nproc_per_node=8 train_retrieval.py \ 
--config ./configs/retrieval_coco.yaml \ 
--output_dir 输出/retrieval_coco

3、算法代码解析

blip_retrieval模型

model = blip_retrieval(pretrained=config['pretrained'], image_size=config['image_size'], vit=config['vit'], 
                             vit_grad_ckpt=config['vit_grad_ckpt'], vit_ckpt_layer=config['vit_ckpt_layer'], 
                             queue_size=config['queue_size'], negative_all_rank=config['negative_all_rank'])

模型训练，，进入到model函数，image【32，3，384，384】

（1）图像编码器（VIT）：提取图片特征

self.visual_encoder, vision_width = create_vit(vit,image_size, vit_grad_ckpt, vit_ckpt_layer)
image_embeds = self.visual_encoder(image)

self.visual_encoder, vision_width = create_vit(vit,image_size, vit_grad_ckpt, vit_ckpt_layer)

进入create_vit函数

def create_vit(vit, image_size, use_grad_checkpointing=False, ckpt_layer=0, drop_path_rate=0):
        
    assert vit in ['base', 'large'], "vit parameter must be base or large"
    if vit=='base':
        vision_width = 768
        visual_encoder = VisionTransformer(img_size=image_size, patch_size=16, embed_dim=vision_width, depth=12, 
                                           num_heads=12, use_grad_checkpointing=use_grad_checkpointing, ckpt_layer=ckpt_layer,
                                           drop_path_rate=0 or drop_path_rate
                                          )   
    elif vit=='large':
        vision_width = 1024
        visual_encoder = VisionTransformer(img_size=image_size, patch_size=16, embed_dim=vision_width, depth=24, 
                                           num_heads=16, use_grad_checkpointing=use_grad_checkpointing, ckpt_layer=ckpt_layer,
                                           drop_path_rate=0.1 or drop_path_rate
                                          )   
    return visual_encoder, vision_width

进入VisionTransformer函数

class VisionTransformer(nn.Module):
    """ Vision Transformer
    A PyTorch impl of : `An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale`  -
        https://arxiv.org/abs/2010.11929
    """
    def __init__(self, img_size=224, patch_size=16, in_chans=3, num_classes=1000, embed_dim=768, depth=12,
                 num_heads=12, mlp_ratio=4., qkv_bias=True, qk_scale=None, representation_size=None,
                 drop_rate=0., attn_drop_rate=0., drop_path_rate=0., norm_layer=None, 
                 use_grad_checkpointing=False, ckpt_layer=0):
        """
        Args:
            img_size (int, tuple): input image size
            patch_size (int, tuple): patch size
            in_chans (int): number of input channels
            num_classes (int): number of classes for classification head
            embed_dim (int): embedding dimension
            depth (int): depth of transformer
            num_heads (int): number of attention heads
            mlp_ratio (int): ratio of mlp hidden dim to embedding dim
            qkv_bias (bool): enable bias for qkv if True
            qk_scale (float): override default qk scale of head_dim ** -0.5 if set
            representation_size (Optional[int]): enable and set representation layer (pre-logits) to this value if set
            drop_rate (float): dropout rate
            attn_drop_rate (float): attention dropout rate
            drop_path_rate (float): stochastic depth rate
            norm_layer: (nn.Module): normalization layer
        """
        super().__init__()
        self.num_features = self.embed_dim = embed_dim  # num_features for consistency with other models
        norm_layer = norm_layer or partial(nn.LayerNorm, eps=1e-6)

        self.patch_embed = PatchEmbed(
            img_size=img_size, patch_size=patch_size, in_chans=in_chans, embed_dim=embed_dim)

        num_patches = self.patch_embed.num_patches

        self.cls_token = nn.Parameter(torch.zeros(1, 1, embed_dim))
        self.pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1, embed_dim))
        self.pos_drop = nn.Dropout(p=drop_rate)

        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, depth)]  # stochastic depth decay rule
        self.blocks = nn.ModuleList([
            Block(
                dim=embed_dim, num_heads=num_heads, mlp_ratio=mlp_ratio, qkv_bias=qkv_bias, qk_scale=qk_scale,
                drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[i], norm_layer=norm_layer,
                use_grad_checkpointing=(use_grad_checkpointing and i>=depth-ckpt_layer)
            )
            for i in range(depth)])
        self.norm = norm_layer(embed_dim)

        trunc_normal_(self.pos_embed, std=.02)
        trunc_normal_(self.cls_token, std=.02)
        self.apply(self._init_weights)

    def _init_weights(self, m):
        if isinstance(m, nn.Linear):
            trunc_normal_(m.weight, std=.02)
            if isinstance(m, nn.Linear) and m.bias is not None:
                nn.init.constant_(m.bias, 0)
        elif isinstance(m, nn.LayerNorm):
            nn.init.constant_(m.bias, 0)
            nn.init.constant_(m.weight, 1.0)

    @torch.jit.ignore
    def no_weight_decay(self):
        return {'pos_embed', 'cls_token'}

    def forward(self, x, register_blk=-1):
        B = x.shape[0]
        x = self.patch_embed(x)

        cls_tokens = self.cls_token.expand(B, -1, -1)  # stole cls_tokens impl from Phil Wang, thanks
        x = torch.cat((cls_tokens, x), dim=1)
  
        x = x + self.pos_embed[:,:x.size(1),:]
        x = self.pos_drop(x)

        for i,blk in enumerate(self.blocks):
            x = blk(x, register_blk==i)
        x = self.norm(x)
        
        return x

    @torch.jit.ignore()
    def load_pretrained(self, checkpoint_path, prefix=''):
        _load_weights(self, checkpoint_path, prefix)
        

ViT架构，如下图所示。ViT（Vision Transformer）是一种基于Transformer的图像分类模型，它将图像看作是一个序列，通过自注意力机制来提取图像的特征表示。ViT的图像特征提取过程可以分为以下几个步骤：1.图像划分：将输入的图像划分为一组固定大小的图像块，每个图像块都被视为一个序列元素。2.嵌入层：对于每个图像块，使用一个嵌入层将其转换为一个向量表示。这个嵌入层通常是一个全连接层或卷积层。3.位置编码：为了保留图像中像素之间的位置信息，ViT使用一种位置编码方法来为每个图像块添加位置信息。这个位置编码通常是一个正弦函数或余弦函数。4.Transformer编码器：为了执行分类，使用额外的[CLS] token来表示全局的图像特征，将嵌入向量和位置编码向量输入到Transformer编码器中，通过多层Transformer编码器来提取图像的特征表示。在每个Transformer编码器中，使用自注意力机制来捕捉图像块之间的关联信息，并使用前馈神经网络来提取高级特征。5.全局池化：最后，将所有Transformer编码器的输出向量进行平均池化或最大池化，得到整个图像的特征表示。通过这种方式，ViT可以将图像转换为一个序列，并使用Transformer编码器来提取图像的特征表示。这种方法不仅可以用于图像分类，还可以用于其他视觉-语言任务，如图像标注、视觉问答等。

1、图像划分和进行位置编码

self.patch_embed = PatchEmbed(
            img_size=img_size, patch_size=patch_size, in_chans=in_chans, embed_dim=embed_dim)

class PatchEmbed(nn.Module):
    """ 2D Image to Patch Embedding
    """
    def __init__(self, img_size=224, patch_size=16, in_chans=3, embed_dim=768, norm_layer=None, flatten=True):
        super().__init__()
        img_size = to_2tuple(img_size)
        patch_size = to_2tuple(patch_size)
        self.img_size = img_size
        self.patch_size = patch_size
        self.grid_size = (img_size[0] // patch_size[0], img_size[1] // patch_size[1])
        self.num_patches = self.grid_size[0] * self.grid_size[1]
        self.flatten = flatten

        self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size)
        self.norm = norm_layer(embed_dim) if norm_layer else nn.Identity()

    def forward(self, x):
        B, C, H, W = x.shape
        assert H == self.img_size[0] and W == self.img_size[1], \
            f"Input image size ({H}*{W}) doesn't match model ({self.img_size[0]}*{self.img_size[1]})."
        x = self.proj(x)
        if self.flatten:
            x = x.flatten(2).transpose(1, 2)  # BCHW -> BNC
        x = self.norm(x)
        return x

2、Transformer编码器

self.blocks = nn.ModuleList([
            Block(
                dim=embed_dim, num_heads=num_heads, mlp_ratio=mlp_ratio, qkv_bias=qkv_bias, qk_scale=qk_scale,
                drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[i], norm_layer=norm_layer,
                use_grad_checkpointing=(use_grad_checkpointing and i>=depth-ckpt_layer)
            )
            for i in range(depth)])#一共12个block

class Block(nn.Module):

    def __init__(self, dim, num_heads, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop=0., attn_drop=0.,
                 drop_path=0., act_layer=nn.GELU, norm_layer=nn.LayerNorm, use_grad_checkpointing=False):
        super().__init__()
        self.norm1 = norm_layer(dim)
        self.attn = Attention(
            dim, num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop)
        # NOTE: drop path for stochastic depth, we shall see if this is better than dropout here
        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
        self.norm2 = norm_layer(dim)
        mlp_hidden_dim = int(dim * mlp_ratio)
        self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)

        if use_grad_checkpointing:
            self.attn = checkpoint_wrapper(self.attn)
            self.mlp = checkpoint_wrapper(self.mlp)

    def forward(self, x, register_hook=False):
        x = x + self.drop_path(self.attn(self.norm1(x), register_hook=register_hook))
        x = x + self.drop_path(self.mlp(self.norm2(x)))
        return x

进行Attention函数

class Attention(nn.Module):
   def __init__(self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0.):
        super().__init__()
        self.num_heads = num_heads #12个头
        head_dim = dim // num_heads #每个头的维度64
        # NOTE scale factor was wrong in my original version, can set manually to be compat with prev weights
        self.scale = qk_scale or head_dim ** -0.5
        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)#对q,k,v同时进行线性投影
        self.attn_drop = nn.Dropout(attn_drop)
        self.proj = nn.Linear(dim, dim)
        self.proj_drop = nn.Dropout(proj_drop)
        self.attn_gradients = None
        self.attention_map = None
        
    def save_attn_gradients(self, attn_gradients):
        self.attn_gradients = attn_gradients
        
    def get_attn_gradients(self):
        return self.attn_gradients
    
    def save_attention_map(self, attention_map):
        self.attention_map = attention_map
        
    def get_attention_map(self):
        return self.attention_map
    
    def forward(self, x, register_hook=False):
        B, N, C = x.shape
        qkv = self.qkv(x).reshape(B, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
        q, k, v = qkv[0], qkv[1], qkv[2]   # make torchscript happy (cannot use tensor as tuple)

        attn = (q @ k.transpose(-2, -1)) * self.scale
        attn = attn.softmax(dim=-1)
        attn = self.attn_drop(attn)
                
        if register_hook:
            self.save_attention_map(attn)
            attn.register_hook(self.save_attn_gradients)        

        x = (attn @ v).transpose(1, 2).reshape(B, N, C)
        x = self.proj(x)
        x = self.proj_drop(x)
        return x

进行入MLP函数

self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)

class Mlp(nn.Module):
    """ MLP as used in Vision Transformer, MLP-Mixer and related networks
    """
    def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.):
        super().__init__()
        out_features = out_features or in_features
        hidden_features = hidden_features or in_features
        self.fc1 = nn.Linear(in_features, hidden_features)
        self.act = act_layer()
        self.fc2 = nn.Linear(hidden_features, out_features)
        self.drop = nn.Dropout(drop)

    def forward(self, x):
        x = self.fc1(x)
        x = self.act(x)
        x = self.drop(x)
        x = self.fc2(x)
        x = self.drop(x)
        return x

（2）文本编码器：提取文本特征

用于为预训练的 BERT 模型创建分词器

self.tokenizer = init_tokenizer() 

def init_tokenizer():
    tokenizer = BertTokenizer.from_pretrained('checkpoints/bert-base-uncased')#下载文件
    tokenizer.add_special_tokens({'bos_token':'[DEC]'})
    tokenizer.add_special_tokens({'additional_special_tokens':['[ENC]']})       
    tokenizer.enc_token_id = tokenizer.additional_special_tokens_ids[0]  
    return tokenizer

self.text_encoder = BertModel(config=med_config, add_pooling_layer=False) 

class BertModel(BertPreTrainedModel):
    """
    The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of
    cross-attention is added between the self-attention layers, following the architecture described in `Attention is
    all you need <https://arxiv.org/abs/1706.03762>`__ by Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit,
    Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin.
    argument and :obj:`add_cross_attention` set to :obj:`True`; an :obj:`encoder_hidden_states` is then expected as an
    input to the forward pass.
    """

    def __init__(self, config, add_pooling_layer=True):
        super().__init__(config)
        self.config = config

        self.embeddings = BertEmbeddings(config)
        
        self.encoder = BertEncoder(config)

        self.pooler = BertPooler(config) if add_pooling_layer else None

        self.init_weights()
 

    def get_input_embeddings(self):
        return self.embeddings.word_embeddings

    def set_input_embeddings(self, value):
        self.embeddings.word_embeddings = value

    def _prune_heads(self, heads_to_prune):
        """
        Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} See base
        class PreTrainedModel
        """
        for layer, heads in heads_to_prune.items():
            self.encoder.layer[layer].attention.prune_heads(heads)

    
    def get_extended_attention_mask(self, attention_mask: Tensor, input_shape: Tuple[int], device: device, is_decoder: bool) -> Tensor:
        """
        Makes broadcastable attention and causal masks so that future and masked tokens are ignored.

        Arguments:
            attention_mask (:obj:`torch.Tensor`):
                Mask with ones indicating tokens to attend to, zeros for tokens to ignore.
            input_shape (:obj:`Tuple[int]`):
                The shape of the input to the model.
            device: (:obj:`torch.device`):
                The device of the input to the model.

        Returns:
            :obj:`torch.Tensor` The extended attention mask, with a the same dtype as :obj:`attention_mask.dtype`.
        """
        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
        # ourselves in which case we just need to make it broadcastable to all heads.
        if attention_mask.dim() == 3:
            extended_attention_mask = attention_mask[:, None, :, :]
        elif attention_mask.dim() == 2:
            # Provided a padding mask of dimensions [batch_size, seq_length]
            # - if the model is a decoder, apply a causal mask in addition to the padding mask
            # - if the model is an encoder, make the mask broadcastable to [batch_size, num_heads, seq_length, seq_length]
            if is_decoder:
                batch_size, seq_length = input_shape

                seq_ids = torch.arange(seq_length, device=device)
                causal_mask = seq_ids[None, None, :].repeat(batch_size, seq_length, 1) <= seq_ids[None, :, None]
                # in case past_key_values are used we need to add a prefix ones mask to the causal mask
                # causal and attention masks must have same type with pytorch version < 1.3
                causal_mask = causal_mask.to(attention_mask.dtype)
   
                if causal_mask.shape[1] < attention_mask.shape[1]:
                    prefix_seq_len = attention_mask.shape[1] - causal_mask.shape[1]
                    causal_mask = torch.cat(
                        [
                            torch.ones((batch_size, seq_length, prefix_seq_len), device=device, dtype=causal_mask.dtype),
                            causal_mask,
                        ],
                        axis=-1,
                    )                     

                extended_attention_mask = causal_mask[:, None, :, :] * attention_mask[:, None, None, :]
            else:
                extended_attention_mask = attention_mask[:, None, None, :]
        else:
            raise ValueError(
                "Wrong shape for input_ids (shape {}) or attention_mask (shape {})".format(
                    input_shape, attention_mask.shape
                )
            )

        # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
        # masked positions, this operation will create a tensor which is 0.0 for
        # positions we want to attend and -10000.0 for masked positions.
        # Since we are adding it to the raw scores before the softmax, this is
        # effectively the same as removing these entirely.
        extended_attention_mask = extended_attention_mask.to(dtype=self.dtype)  # fp16 compatibility
        extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0
        return extended_attention_mask
    
    def forward(
        self,
        input_ids=None,
        attention_mask=None,
        position_ids=None,
        head_mask=None,
        inputs_embeds=None,
        encoder_embeds=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        past_key_values=None,
        use_cache=None,
        output_attentions=None,
        output_hidden_states=None,
        return_dict=None,
        is_decoder=False,
        mode='multimodal',
    ):
        r"""
        encoder_hidden_states  (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
            Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
            the model is configured as a decoder.
        encoder_attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
            Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in
            the cross-attention if the model is configured as a decoder. Mask values selected in ``[0, 1]``:
            - 1 for tokens that are **not masked**,
            - 0 for tokens that are **masked**.
        past_key_values (:obj:`tuple(tuple(torch.FloatTensor))` of length :obj:`config.n_layers` with each tuple having 4 tensors of shape :obj:`(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`):
            Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding.
            If :obj:`past_key_values` are used, the user can optionally input only the last :obj:`decoder_input_ids`
            (those that don't have their past key value states given to this model) of shape :obj:`(batch_size, 1)`
            instead of all :obj:`decoder_input_ids` of shape :obj:`(batch_size, sequence_length)`.
        use_cache (:obj:`bool`, `optional`):
            If set to :obj:`True`, :obj:`past_key_values` key value states are returned and can be used to speed up
            decoding (see :obj:`past_key_values`).
        """
        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
        )
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        if is_decoder:
            use_cache = use_cache if use_cache is not None else self.config.use_cache
        else:
            use_cache = False

        if input_ids is not None and inputs_embeds is not None:
            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
        elif input_ids is not None:
            input_shape = input_ids.size()
            batch_size, seq_length = input_shape
            device = input_ids.device
        elif inputs_embeds is not None:
            input_shape = inputs_embeds.size()[:-1]
            batch_size, seq_length = input_shape
            device = inputs_embeds.device
        elif encoder_embeds is not None:    
            input_shape = encoder_embeds.size()[:-1]
            batch_size, seq_length = input_shape 
            device = encoder_embeds.device
        else:
            raise ValueError("You have to specify either input_ids or inputs_embeds or encoder_embeds")

        # past_key_values_length
        past_key_values_length = past_key_values[0][0].shape[2] if past_key_values is not None else 0

        if attention_mask is None:
            attention_mask = torch.ones(((batch_size, seq_length + past_key_values_length)), device=device)
            
        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
        # ourselves in which case we just need to make it broadcastable to all heads.
        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape, 
                                                                                 device, is_decoder)

        # If a 2D or 3D attention mask is provided for the cross-attention
        # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
        if encoder_hidden_states is not None:
            if type(encoder_hidden_states) == list:
                encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states[0].size()
            else:
                encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states.size()
            encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)
            
            if type(encoder_attention_mask) == list:
                encoder_extended_attention_mask = [self.invert_attention_mask(mask) for mask in encoder_attention_mask]
            elif encoder_attention_mask is None:
                encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device)
                encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
            else:    
                encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
        else:
            encoder_extended_attention_mask = None

        # Prepare head mask if needed
        # 1.0 in head_mask indicate we keep the head
        # attention_probs has shape bsz x n_heads x N x N
        # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
        # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
        head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers)
        
        if encoder_embeds is None:
            embedding_output = self.embeddings(
                input_ids=input_ids,
                position_ids=position_ids,
                inputs_embeds=inputs_embeds,
                past_key_values_length=past_key_values_length,
            )
        else:
            embedding_output = encoder_embeds
            
        encoder_outputs = self.encoder(
            embedding_output,
            attention_mask=extended_attention_mask,
            head_mask=head_mask,
            encoder_hidden_states=encoder_hidden_states,
            encoder_attention_mask=encoder_extended_attention_mask,
            past_key_values=past_key_values,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
            mode=mode,
        )
        sequence_output = encoder_outputs[0]
        pooled_output = self.pooler(sequence_output) if self.pooler is not None else None

        if not return_dict:
            return (sequence_output, pooled_output) + encoder_outputs[1:]

        return BaseModelOutputWithPoolingAndCrossAttentions(
            last_hidden_state=sequence_output,
            pooler_output=pooled_output,
            past_key_values=encoder_outputs.past_key_values,
            hidden_states=encoder_outputs.hidden_states,
            attentions=encoder_outputs.attentions,
            cross_attentions=encoder_outputs.cross_attentions,
        )

1、从单词和位置嵌入构建嵌入

class BertEmbeddings(nn.Module):
    """Construct the embeddings from word and position embeddings."""

    def __init__(self, config):
        super().__init__()
        self.word_embeddings = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=config.pad_token_id)
        self.position_embeddings = nn.Embedding(config.max_position_embeddings, config.hidden_size)

        # self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load
        # any TensorFlow checkpoint file
        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)

        # position_ids (1, len position emb) is contiguous in memory and exported when serialized
        self.register_buffer("position_ids", torch.arange(config.max_position_embeddings).expand((1, -1)))
        self.position_embedding_type = getattr(config, "position_embedding_type", "absolute")
        
        self.config = config

    def forward(
        self, input_ids=None, position_ids=None, inputs_embeds=None, past_key_values_length=0
    ):
        if input_ids is not None:
            input_shape = input_ids.size()
        else:
            input_shape = inputs_embeds.size()[:-1]

        seq_length = input_shape[1]

        if position_ids is None:
            position_ids = self.position_ids[:, past_key_values_length : seq_length + past_key_values_length]

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

        embeddings = inputs_embeds

        if self.position_embedding_type == "absolute":
            position_embeddings = self.position_embeddings(position_ids)
            embeddings += position_embeddings
        embeddings = self.LayerNorm(embeddings)
        embeddings = self.dropout(embeddings)
        return embeddings

2、BertEncoder

class BertEncoder(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.config = config
        self.layer = nn.ModuleList([BertLayer(config,i) for i in range(config.num_hidden_layers)])
        self.gradient_checkpointing = False

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        head_mask=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        past_key_values=None,
        use_cache=None,
        output_attentions=False,
        output_hidden_states=False,
        return_dict=True,
        mode='multimodal',
    ):
        all_hidden_states = () if output_hidden_states else None
        all_self_attentions = () if output_attentions else None
        all_cross_attentions = () if output_attentions and self.config.add_cross_attention else None

        next_decoder_cache = () if use_cache else None
               
        for i in range(self.config.num_hidden_layers):
            layer_module = self.layer[i]
            if output_hidden_states:
                all_hidden_states = all_hidden_states + (hidden_states,)

            layer_head_mask = head_mask[i] if head_mask is not None else None
            past_key_value = past_key_values[i] if past_key_values is not None else None

            if self.gradient_checkpointing and self.training:

                if use_cache:
                    logger.warn(
                        "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                    )
                    use_cache = False

                def create_custom_forward(module):
                    def custom_forward(*inputs):
                        return module(*inputs, past_key_value, output_attentions)

                    return custom_forward

                layer_outputs = torch.utils.checkpoint.checkpoint(
                    create_custom_forward(layer_module),
                    hidden_states,
                    attention_mask,
                    layer_head_mask,
                    encoder_hidden_states,
                    encoder_attention_mask,
                    mode=mode,
                )
            else:
                layer_outputs = layer_module(
                    hidden_states,
                    attention_mask,
                    layer_head_mask,
                    encoder_hidden_states,
                    encoder_attention_mask,
                    past_key_value,
                    output_attentions,
                    mode=mode,
                )

            hidden_states = layer_outputs[0]
            if use_cache:
                next_decoder_cache += (layer_outputs[-1],)
            if output_attentions:
                all_self_attentions = all_self_attentions + (layer_outputs[1],)

        if output_hidden_states:
            all_hidden_states = all_hidden_states + (hidden_states,)

        if not return_dict:
            return tuple(
                v
                for v in [
                    hidden_states,
                    next_decoder_cache,
                    all_hidden_states,
                    all_self_attentions,
                    all_cross_attentions,
                ]
                if v is not None
            )
        return BaseModelOutputWithPastAndCrossAttentions(
            last_hidden_state=hidden_states,
            past_key_values=next_decoder_cache,
            hidden_states=all_hidden_states,
            attentions=all_self_attentions,
            cross_attentions=all_cross_attentions,
        )

进入BertLayer

class BertLayer(nn.Module):
    def __init__(self, config, layer_num):
        super().__init__()
        self.config = config
        self.chunk_size_feed_forward = config.chunk_size_feed_forward
        self.seq_len_dim = 1
        self.attention = BertAttention(config)      
        self.layer_num = layer_num          
        if self.config.add_cross_attention:
            self.crossattention = BertAttention(config, is_cross_attention=self.config.add_cross_attention)
        self.intermediate = BertIntermediate(config)
        self.output = BertOutput(config)

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        head_mask=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        past_key_value=None,
        output_attentions=False,
        mode=None,
    ):
        # decoder uni-directional self-attention cached key/values tuple is at positions 1,2
        self_attn_past_key_value = past_key_value[:2] if past_key_value is not None else None
        self_attention_outputs = self.attention(
            hidden_states,
            attention_mask,
            head_mask,
            output_attentions=output_attentions,
            past_key_value=self_attn_past_key_value,
        )
        attention_output = self_attention_outputs[0]

        outputs = self_attention_outputs[1:-1]
        present_key_value = self_attention_outputs[-1]

        if mode=='multimodal':
            assert encoder_hidden_states is not None, "encoder_hidden_states must be given for cross-attention layers"

            cross_attention_outputs = self.crossattention(
                attention_output,
                attention_mask,
                head_mask,
                encoder_hidden_states,
                encoder_attention_mask,
                output_attentions=output_attentions,
            )
            attention_output = cross_attention_outputs[0]
            outputs = outputs + cross_attention_outputs[1:-1]  # add cross attentions if we output attention weights                               
        layer_output = apply_chunking_to_forward(
            self.feed_forward_chunk, self.chunk_size_feed_forward, self.seq_len_dim, attention_output
        )
        outputs = (layer_output,) + outputs

        outputs = outputs + (present_key_value,)

        return outputs

    def feed_forward_chunk(self, attention_output):
        intermediate_output = self.intermediate(attention_output)
        layer_output = self.output(intermediate_output, attention_output)
        return layer_output

进入BertAttention

class BertAttention(nn.Module):
    def __init__(self, config, is_cross_attention=False):
        super().__init__()
        self.self = BertSelfAttention(config, is_cross_attention)
        self.output = BertSelfOutput(config)
        self.pruned_heads = set()

    def prune_heads(self, heads):
        if len(heads) == 0:
            return
        heads, index = find_pruneable_heads_and_indices(
            heads, self.self.num_attention_heads, self.self.attention_head_size, self.pruned_heads
        )

        # Prune linear layers
        self.self.query = prune_linear_layer(self.self.query, index)
        self.self.key = prune_linear_layer(self.self.key, index)
        self.self.value = prune_linear_layer(self.self.value, index)
        self.output.dense = prune_linear_layer(self.output.dense, index, dim=1)

        # Update hyper params and store pruned heads
        self.self.num_attention_heads = self.self.num_attention_heads - len(heads)
        self.self.all_head_size = self.self.attention_head_size * self.self.num_attention_heads
        self.pruned_heads = self.pruned_heads.union(heads)

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        head_mask=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        past_key_value=None,
        output_attentions=False,
    ):
        self_outputs = self.self(
            hidden_states,
            attention_mask,
            head_mask,
            encoder_hidden_states,
            encoder_attention_mask,
            past_key_value,
            output_attentions,
        )
        attention_output = self.output(self_outputs[0], hidden_states)
        outputs = (attention_output,) + self_outputs[1:]  # add attentions if we output them
        return outputs

进入BertSelfAttention

class BertSelfAttention(nn.Module):
    def __init__(self, config, is_cross_attention):
        super().__init__()
        self.config = config
        if config.hidden_size % config.num_attention_heads != 0 and not hasattr(config, "embedding_size"):
            raise ValueError(
                "The hidden size (%d) is not a multiple of the number of attention "
                "heads (%d)" % (config.hidden_size, config.num_attention_heads)
            )
        
        self.num_attention_heads = config.num_attention_heads
        self.attention_head_size = int(config.hidden_size / config.num_attention_heads)
        self.all_head_size = self.num_attention_heads * self.attention_head_size

        self.query = nn.Linear(config.hidden_size, self.all_head_size)
        if is_cross_attention:
            self.key = nn.Linear(config.encoder_width, self.all_head_size)
            self.value = nn.Linear(config.encoder_width, self.all_head_size)
        else:
            self.key = nn.Linear(config.hidden_size, self.all_head_size)
            self.value = nn.Linear(config.hidden_size, self.all_head_size)

        self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
        self.position_embedding_type = getattr(config, "position_embedding_type", "absolute")
        if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
            self.max_position_embeddings = config.max_position_embeddings
            self.distance_embedding = nn.Embedding(2 * config.max_position_embeddings - 1, self.attention_head_size)
        self.save_attention = False   
            
    def save_attn_gradients(self, attn_gradients):
        self.attn_gradients = attn_gradients
        
    def get_attn_gradients(self):
        return self.attn_gradients
    
    def save_attention_map(self, attention_map):
        self.attention_map = attention_map
        
    def get_attention_map(self):
        return self.attention_map
    
    def transpose_for_scores(self, x):
        new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size)
        x = x.view(*new_x_shape)
        return x.permute(0, 2, 1, 3)

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        head_mask=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        past_key_value=None,
        output_attentions=False,
    ):
        mixed_query_layer = self.query(hidden_states)

        # If this is instantiated as a cross-attention module, the keys
        # and values come from an encoder; the attention mask needs to be
        # such that the encoder's padding tokens are not attended to.
        is_cross_attention = encoder_hidden_states is not None

        if is_cross_attention:
            key_layer = self.transpose_for_scores(self.key(encoder_hidden_states))
            value_layer = self.transpose_for_scores(self.value(encoder_hidden_states))
            attention_mask = encoder_attention_mask
        elif past_key_value is not None:
            key_layer = self.transpose_for_scores(self.key(hidden_states))
            value_layer = self.transpose_for_scores(self.value(hidden_states))
            key_layer = torch.cat([past_key_value[0], key_layer], dim=2)
            value_layer = torch.cat([past_key_value[1], value_layer], dim=2)
        else:
            key_layer = self.transpose_for_scores(self.key(hidden_states))
            value_layer = self.transpose_for_scores(self.value(hidden_states))

        query_layer = self.transpose_for_scores(mixed_query_layer)

        past_key_value = (key_layer, value_layer)

        # Take the dot product between "query" and "key" to get the raw attention scores.
        attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))

        if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
            seq_length = hidden_states.size()[1]
            position_ids_l = torch.arange(seq_length, dtype=torch.long, device=hidden_states.device).view(-1, 1)
            position_ids_r = torch.arange(seq_length, dtype=torch.long, device=hidden_states.device).view(1, -1)
            distance = position_ids_l - position_ids_r
            positional_embedding = self.distance_embedding(distance + self.max_position_embeddings - 1)
            positional_embedding = positional_embedding.to(dtype=query_layer.dtype)  # fp16 compatibility

            if self.position_embedding_type == "relative_key":
                relative_position_scores = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
                attention_scores = attention_scores + relative_position_scores
            elif self.position_embedding_type == "relative_key_query":
                relative_position_scores_query = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
                relative_position_scores_key = torch.einsum("bhrd,lrd->bhlr", key_layer, positional_embedding)
                attention_scores = attention_scores + relative_position_scores_query + relative_position_scores_key

        attention_scores = attention_scores / math.sqrt(self.attention_head_size)
        if attention_mask is not None:
            # Apply the attention mask is (precomputed for all layers in BertModel forward() function)
            attention_scores = attention_scores + attention_mask

        # Normalize the attention scores to probabilities.
        attention_probs = nn.Softmax(dim=-1)(attention_scores)
        
        if is_cross_attention and self.save_attention:
            self.save_attention_map(attention_probs)
            attention_probs.register_hook(self.save_attn_gradients)         

        # This is actually dropping out entire tokens to attend to, which might
        # seem a bit unusual, but is taken from the original Transformer paper.
        attention_probs_dropped = self.dropout(attention_probs)

        # Mask heads if we want to
        if head_mask is not None:
            attention_probs_dropped = attention_probs_dropped * head_mask

        context_layer = torch.matmul(attention_probs_dropped, value_layer)

        context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
        new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)
        context_layer = context_layer.view(*new_context_layer_shape)

        outputs = (context_layer, attention_probs) if output_attentions else (context_layer,)

        outputs = outputs + (past_key_value,)
        return outputs

进入BertSelfOutput函数

进入BertIntermediate函数

loss计算

Image-Text Contrastive Loss (ITC)是BLIP模型中用于对齐视觉和语言表示的损失函数之一。

下面是ITC损失的计算过程：

1. 首先，对于每个训练样本，包括一个图像和一个文本描述，通过模型的视觉和语言编码器分别得到它们的表示。

2. 对于图像和文本的表示，分别计算它们的L2范数，并对它们进行归一化，以确保它们在相同的空间中。

3. 然后，计算图像和文本表示的内积，得到它们之间的相似度分数。

4. 接下来，对于每个样本，将其分为正样本对和负样本对。正样本对是指图像和文本描述是匹配的，而负样本对是指图像和文本描述不匹配。

5. 对于每个正样本对，计算图像表示和文本表示之间的相似度分数，并将其作为正样本的得分。

6. 对于每个负样本对，同样计算图像表示和文本表示之间的相似度分数，并将其作为负样本的得分。

7. 最后，通过对比正样本对和负样本对的得分，使用交叉熵损失函数来最小化它们之间的差异，以促进正样本对的相似度得分更高，负样本对的相似度得分更低。

通过这种方式，ITC损失函数可以帮助模型学习到视觉和语言表示之间的对齐关系，从而在视觉-语言任务中取得更好的性能。

Image-Text Matching Loss (ITM)是BLIP模型中用于训练图像-文本匹配的损失函数之一。

下面是ITM损失的计算过程：

1. 对于每个训练样本，包括一个图像和一个文本描述，通过模型的视觉编码器和文本编码器分别得到它们的表示。

2. 对于图像和文本的表示，分别计算它们的L2范数，并对它们进行归一化，以确保它们在相同的空间中。

3. 然后，将图像表示和文本表示拼接在一起，得到一个多模态表示。

4. 对于每个样本，将其分为正样本对和负样本对。正样本对是指图像和文本描述是匹配的，而负样本对是指图像和文本描述不匹配。

5. 对于每个样本，使用一个线性层（ITM head）来预测它是正样本对还是负样本对。这可以通过将多模态表示输入到ITM head中，得到一个二元分类的概率分数来实现。

6. 最后，通过对比预测的概率分数和真实标签（1表示正样本对，0表示负样本对），使用二元交叉熵损失函数来最小化它们之间的差异，以促进正样本对的概率分数更高，负样本对的概率分数更低。

通过这种方式，ITM损失函数可以帮助模型学习到图像和文本表示之间的匹配关系，从而在图像-文本匹配任务中取得更好的性能。

4、算法问题

用于图像检索和微调的特征提取 #25

针对没有注释的自定义数据集微调 BLIP 图像检索 第55章

使用一个 GPU 进行训练 第43章

使用一个 GPU 进行训练 第43章