BLIP 算法阅读记录---一个许多多模态大语言模型的基本组件

论文地址：😈

一、环境配置以及数据集准备

数据集准备

官网提供了下载数据集json文件的接口。但是很可能打不开，因为其放在了谷歌云上

https://storage.googleapis.com/

不过不要担心，网页打不开，咱们可以利用python去爬它。

# urlopen模块 读取数据
from __future__ import (absolute_import, division, print_function, unicode_literals)

from urllib.request import urlopen
import json

json_url = 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/datasets/ccs_filtered.json'

response = urlopen(json_url)
# 读取数据
req = response.read()
# 写入文件
with open('ccs_filtered.json', 'wb') as f:  # 保存的路径自行设置
    f.write(req)
f.close()
# 加载json格式
# file_urllid = json.loads(req)
# print(file_urllid)

数据集格式展示

正如官网所述，json文件中的格式为{'image': path_of_image, 'caption': text_of_image}，部分展示如下：

[{"caption": "a wooden chair in the living room", "url": "http://static.flickr.com/2723/4385058960_b0f291553e.jpg"},
  {"caption": "the plate near the door almost makes it look like the cctv camera is the logo of the hotel", "url": "http://static.flickr.com/3348/3183472382_156bcbc461.jpg"},
  {"caption": "a close look at winter stoneflies reveals mottled wings and black or brown bodies photo by jason du pont", "url": "http://static.flickr.com/114/291586170_82b2c80750.jpg"},
  {"caption": "karipol leipzig, germany, abandoned factory for house und car cleaning supplies in leipzig founded in 1897, closed in 1995", "url": "http://static.flickr.com/5170/5208095391_f030f31dd1.jpg"},
  {"caption": "this train car is parked permanently in a yard in royal, nebraska", "url": "http://static.flickr.com/3067/4554530891_f0d45b0967.jpg"},
  {"caption": "another shot of the building next to the sixth floor museum in dallas", "url": "http://static.flickr.com/3145/2858049135_0944b40a34.jpg"},
  {"caption": "a tropical forest in the train station how cool is that", "url": "http://static.flickr.com/4113/5212209590_6c7d40a7f6.jpg"},
  {"caption": "le hast le hill in le pink le shirt", "url": "http://static.flickr.com/1338/4597496270_7d91f01f88.jpg"},
  {"caption": "daniel up by the red bag tackling", "url": "http://static.flickr.com/3129/2783268248_4930260b9d.jpg"},
  {"caption": "a window on a service building is seen at allerton park tuesday, oct 27, 2009, in monticello, ill", "url": "http://static.flickr.com/2516/4054292066_3b74bfcf89.jpg"},
  {"caption": "girls jumping off of the 50ft cliff at abique lake in santa fe new mexico", "url": "http://static.flickr.com/2637/4371962912_77b4c74878.jpg"},
  {"caption": "a young chhantyal girl in traditional dress", "url": "http://static.flickr.com/3075/3240630067_772ebf5504.jpg"},
  {"caption": "view from the train to snowdon hidden in the cloud", "url": "http://static.flickr.com/4114/4803115610_711924e2ab.jpg"},
  {"caption": "the only non white tiger in the habitat when i went there", "url": "http://static.flickr.com/2481/3873285144_25aa3013ca.jpg"},
  {"caption": "the contrast of the flowering pear tree against the bare branches of the other trees caught my eye", "url": "http://static.flickr.com/3198/3124550410_b70442da56.jpg"},
  {"caption": "this is my friend taking a nap in my sleeping bag with our friend's dog for company", "url": "http://static.flickr.com/1167/1466307446_c1a332c5ec.jpg"},
  {"caption": "view of old castle in field ii", "url": "http://static.flickr.com/4101/4758652582_8a1d44a1a0.jpg"},

a wooden chair in the living room

this is my friend taking a nap in my sleeping bag with our friend's dog for company

环境配置，按照官网所述即可

# 不是顺序执行，大致意思
conda create -n BLIP python=3.8
pip install -r requirement.txt
pip install opencv-python
pip install opencv-python-headless

按照官网，去configs/pretrain.yaml修改json files的路径。然后自行创建保存的路径

mkdir output/pretrain  (大致意思，不是执行这个)

执行测试，（这里只用了一块GPU）

python -m torch.distributed.run --nproc_per_node=1 pretrain.py --config ./configs/Pretrain.yaml --output_dir output/Pretrain

出现问题

ModuleNotFoundError: No module named 'ruamel_yaml'

解决方法

# 将
import ruamel_yaml as yaml
#改为
import yaml

出现问题

FileNotFoundError: [Errno 2] No such file or directory: './configs/Pretrain.yaml'

解决方法

用绝对路径代替

出现问题

RuntimeError: The NVIDIA driver on your system is too old (found version 11060). Please update your GPU driver by downloading and installing a new version from the URL: http://www.nvidia.com/Download/index.aspx Alternatively, go to: https://pytorch.org to install a PyTorch version that has been compiled with your version of the CUDA driver.

解决方法

见这里 😼

出现问题

FileNotFoundError: [Errno 2] No such file or directory: 'configs/bert_config.json'

解决办法

用绝对路劲代替

出现问题

TypeError: '<=' not supported between instances of 'float' and 'str'

解决办法

yaml错把3e-4等识别为str型数据

lr=float(config['init_lr']) 转成float型，或者去yaml文件中将

3e-4

1e-6

等直接换成

0.0003

0.0000001

出现问题

image = Image.open(ann['image']).convert('RGB')
KeyError: 'image'

解决办法

见本文中的第二部分中的读入网络图片的调整

二、一些调整

vit_base的预训练模型

执行训练时会去网上下载这个预训练模型。为了避免重复下载，可将下载的预训练模型保存到一个文件夹中，并更改代码中的设置

checkpoint = torch.load('******/deit_base_patch16_224-b5f2ef4d.pth', ap_location="cpu")

远程debug的设置

遇见的问题

error:unrecognized argument: --local-rank=0

解决方法见这里 🐰

Tokenizer初始化失败

由于初始化需要去hugging face下载模型文，所以有可能由于网络原因报错

TimeoutError: [Errno 110] Connection timed out

解决方法见这里😼

并且 self.text_encoder 和self.text_decoder也需要同步进行更改

读入网络图片的调整

原数据集json文件太大，所以我编写脚本截取了一部分，只有4M的，20000个图片。

我不知道为什么，官网说的数据集json文件格式是{'image': path_of_image, 'caption': text_of_image}，但是我下载的json文件格式是{'url': url_of_image, 'caption': text_of_image}。这个可能是我下载的是CC3M+CC12M+SBU，来自网页的，如果是coco的话可能就不一样了，就和他所述的格式一样。

所以，为了在网页上抓取并读入图片，需要修改一下源代码（参考了这里🐰）

pretrain_dataset.py 中

# 导入库
from skimage import io

# 插入代码
self.use_url = True

# 以及函数
    def read_image_from_url(self, url):
        # 从URL下载图像
        image = io.imread(url)
        image = Image.fromarray(image)
        return image
# 将源代码修改为

    def __getitem__(self, index):    
        ann = self.annotation[index]   
        if self.use_url:
            image = self.read_image_from_url(ann['url'])
        else:
            image = Image.open(ann['image']).convert('RGB')
        image = self.transform(image)
        caption = pre_caption(ann['caption'],30)
        
        return image, caption

然而可能遇到网络原因，403，不能读取图片。因此由于本次只是测试，就写了个脚本下载了截取的一些图片并配置json格式到 {'image': path_of_image, 'caption': text_of_image}。具体地，截取了原数据集json文件的前300个，

import json

path='/root/data/zjx/Code-subject/BLIP/ccs_filtered.json'
save_path ='/root/data/zjx/Code-subject/BLIP/cut_css_filtered.json'

data = json.load(open(path, 'r'))
with open(save_path, 'w') as f2:
    save_len = 300
    for i in range(save_len):
        js = json.dumps(data[i])
        if i==0:
            f2.write('['+js+',')
        elif i == save_len-1:
            f2.write(js+']')
        else:
            f2.write(js+',')
print('finish')

下载图片保存到文件夹并实现配置json文件

import json
from skimage import io
from tqdm import tqdm


def download_from_image(url, save_path, ind):
    image = io.imread(url)
    if ind <10:
        index_ = '000'+ str(ind)
    elif 9<ind<100:
        index_ = '00'+str(ind)
    else:
        index_ = '0'+str(ind)
    io.imsave(save_path+'/'+index_+'.jpg', image)
    return index_+'.jpg'

def do_this_work(ori_json_path, new_json_path, yunduan_path, save_image_path, save_image_path_name):
    dict_data = json.load(open(ori_json_path, 'r'))
    num = len(dict_data)
    with open(new_json_path, 'w') as f2:
        for i, dict_info in enumerate(tqdm(dict_data)):
            new_dict = {}
            new_dict["caption"]=dict_info["caption"]
            image_path = download_from_image(dict_info['url'], save_image_path, i)
            new_dict["image"] = yunduan_path+'/'+save_image_path_name+'/'+image_path
            new_dict = json.dumps(new_dict)
            if i == 0:
                f2.write('[' + new_dict + ',')
            elif i == num - 1:
                f2.write(new_dict + ']')
            else:
                f2.write(new_dict + ',')

if __name__=='__main__':
    path = './cut_css_filtered.json'
    save_image_path = './save_imgae'
    save_image_path_name = 'save_imgae'
    save_new_json_path = './new_ccs_filtered.json'
    yunduan_path = '/root/data/zjx/Code-subject/BLIP'
    do_this_work(path, save_new_json_path, yunduan_path, save_image_path, save_image_path_name)

接着把yaml文件中的batch size 调小了

本地IDE debug 的配置

--nproc_per_node=1
--use_env
/root/data/zjx/Code-subject/BLIP/BLIP-main/pretrain.py
--config
/root/data/zjx/Code-subject/BLIP/BLIP-main/configs/pretrain.yaml
--output_dir
output/Pretrain

映射关系

/root/anaconda3/envs/BLIP/lib/python3.8/site-packages/torch/distributed/launch.py

C:\Users\Lenovo\AppData\Local\JetBrains\PyCharm2021.2\remote_sources\2036786058\597056065\torch\distributed\launch.py

三、训练过程

Image Encoder Layer

采用的ViT架构，大致流程如下

接着用线性层映射了一下，并进行了二范数归一化。

Text Encoder

首先对caption进行量化，返回的是个BtachEncoding，其中的属性如下举例，编码的最大长度为本次选取的为30

{'input_ids': tensor([[  101,  1037,  2485,  2298,  2012,  3467,  2962, 24019,  7657,  9587,
         26328,  2094,  4777,  1998,  2304,  2030,  2829,  4230,  6302,  2011,
          4463,  4241, 21179,   102,     0,     0,     0,     0,     0,     0],
        [  101,  1037,  4799,  3242,  1999,  1996,  2542,  2282,   102,     0,
             0,     0,     0,     0,     0,     0,     0,     0,     0,     0,
             0,     0,     0,     0,     0,     0,     0,     0,     0,     0]],
       device='cuda:0'), 'token_type_ids': tensor([[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
         0, 0, 0, 0, 0, 0],
        [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
         0, 0, 0, 0, 0, 0]], device='cuda:0'), 'attention_mask': tensor([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
         0, 0, 0, 0, 0, 0],
        [1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
         0, 0, 0, 0, 0, 0]], device='cuda:0')}

mask的长度和句子量化后的实值长度一样。

文本Encoder采用的BertModel结构

对于mask的处理

1、get_extended_attention_mask

最终extended_attention_mask 为 None

2、get_head_mask

最终 [None, None, None, None, None, None, None, None, None, None, None, None]

其长度等于隐藏层的数量 12

大致流程如下

网络结构

BLIP_Pretrain(
  (visual_encoder): VisionTransformer(
    (patch_embed): PatchEmbed(
      (proj): Conv2d(3, 768, kernel_size=(16, 16), stride=(16, 16))
      (norm): Identity()
    )
    (pos_drop): Dropout(p=0.0, inplace=False)
    (blocks): ModuleList(
      (0): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
      (1): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
      (2): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
      (3): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
      (4): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
      (5): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
      (6): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
      (7): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
      (8): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
      (9): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
      (10): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
      (11): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
    )
    (norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
  )
  (text_encoder): BertModel(
    (embeddings): BertEmbeddings(
      (word_embeddings): Embedding(30524, 768)
      (position_embeddings): Embedding(512, 768)
      (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
      (dropout): Dropout(p=0.1, inplace=False)
    )
    (encoder): BertEncoder(
      (layer): ModuleList(
        (0): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (1): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (2): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (3): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (4): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (5): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (6): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (7): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (8): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (9): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (10): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (11): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
      )
    )
  )
  (vision_proj): Linear(in_features=768, out_features=256, bias=True)
  (text_proj): Linear(in_features=768, out_features=256, bias=True)
  (itm_head): Linear(in_features=768, out_features=2, bias=True)
  (visual_encoder_m): VisionTransformer(
    (patch_embed): PatchEmbed(
      (proj): Conv2d(3, 768, kernel_size=(16, 16), stride=(16, 16))
      (norm): Identity()
    )
    (pos_drop): Dropout(p=0.0, inplace=False)
    (blocks): ModuleList(
      (0): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
      (1): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
      (2): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
      (3): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
      (4): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
      (5): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
      (6): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
      (7): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
      (8): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
      (9): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
      (10): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
      (11): Block(
        (norm1): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (attn): Attention(
          (qkv): Linear(in_features=768, out_features=2304, bias=True)
          (attn_drop): Dropout(p=0.0, inplace=False)
          (proj): Linear(in_features=768, out_features=768, bias=True)
          (proj_drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): Identity()
        (norm2): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=768, out_features=3072, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=3072, out_features=768, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
      )
    )
    (norm): LayerNorm((768,), eps=1e-06, elementwise_affine=True)
  )
  (vision_proj_m): Linear(in_features=768, out_features=256, bias=True)
  (text_encoder_m): BertModel(
    (embeddings): BertEmbeddings(
      (word_embeddings): Embedding(30524, 768, padding_idx=0)
      (position_embeddings): Embedding(512, 768)
      (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
      (dropout): Dropout(p=0.1, inplace=False)
    )
    (encoder): BertEncoder(
      (layer): ModuleList(
        (0): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (1): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (2): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (3): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (4): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (5): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (6): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (7): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (8): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (9): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (10): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (11): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
      )
    )
  )
  (text_proj_m): Linear(in_features=768, out_features=256, bias=True)
  (text_decoder): BertLMHeadModel(
    (bert): BertModel(
      (embeddings): BertEmbeddings(
        (word_embeddings): Embedding(30524, 768)
        (position_embeddings): Embedding(512, 768)
        (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
        (dropout): Dropout(p=0.1, inplace=False)
      )
      (encoder): BertEncoder(
        (layer): ModuleList(
          (0): BertLayer(
            (attention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (crossattention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (intermediate): BertIntermediate(
              (dense): Linear(in_features=768, out_features=3072, bias=True)
            )
            (output): BertOutput(
              (dense): Linear(in_features=3072, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (1): BertLayer(
            (attention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (crossattention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (intermediate): BertIntermediate(
              (dense): Linear(in_features=768, out_features=3072, bias=True)
            )
            (output): BertOutput(
              (dense): Linear(in_features=3072, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (2): BertLayer(
            (attention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (crossattention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (intermediate): BertIntermediate(
              (dense): Linear(in_features=768, out_features=3072, bias=True)
            )
            (output): BertOutput(
              (dense): Linear(in_features=3072, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (3): BertLayer(
            (attention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (crossattention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (intermediate): BertIntermediate(
              (dense): Linear(in_features=768, out_features=3072, bias=True)
            )
            (output): BertOutput(
              (dense): Linear(in_features=3072, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (4): BertLayer(
            (attention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (crossattention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (intermediate): BertIntermediate(
              (dense): Linear(in_features=768, out_features=3072, bias=True)
            )
            (output): BertOutput(
              (dense): Linear(in_features=3072, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (5): BertLayer(
            (attention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (crossattention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (intermediate): BertIntermediate(
              (dense): Linear(in_features=768, out_features=3072, bias=True)
            )
            (output): BertOutput(
              (dense): Linear(in_features=3072, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (6): BertLayer(
            (attention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (crossattention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (intermediate): BertIntermediate(
              (dense): Linear(in_features=768, out_features=3072, bias=True)
            )
            (output): BertOutput(
              (dense): Linear(in_features=3072, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (7): BertLayer(
            (attention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (crossattention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (intermediate): BertIntermediate(
              (dense): Linear(in_features=768, out_features=3072, bias=True)
            )
            (output): BertOutput(
              (dense): Linear(in_features=3072, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (8): BertLayer(
            (attention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (crossattention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (intermediate): BertIntermediate(
              (dense): Linear(in_features=768, out_features=3072, bias=True)
            )
            (output): BertOutput(
              (dense): Linear(in_features=3072, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (9): BertLayer(
            (attention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (crossattention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (intermediate): BertIntermediate(
              (dense): Linear(in_features=768, out_features=3072, bias=True)
            )
            (output): BertOutput(
              (dense): Linear(in_features=3072, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (10): BertLayer(
            (attention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (crossattention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (intermediate): BertIntermediate(
              (dense): Linear(in_features=768, out_features=3072, bias=True)
            )
            (output): BertOutput(
              (dense): Linear(in_features=3072, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (11): BertLayer(
            (attention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (crossattention): BertAttention(
              (self): BertSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (intermediate): BertIntermediate(
              (dense): Linear(in_features=768, out_features=3072, bias=True)
            )
            (output): BertOutput(
              (dense): Linear(in_features=3072, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
        )
      )
    )
    (cls): BertOnlyMLMHead(
      (predictions): BertLMPredictionHead(
        (transform): BertPredictionHeadTransform(
          (dense): Linear(in_features=768, out_features=768, bias=True)
          (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
        )
        (decoder): Linear(in_features=768, out_features=30524, bias=True)
      )
    )
  )
)