clip_encoder.py

import torch
import torch.nn as nn

from transformers import CLIPVisionModel, CLIPImageProcessor, CLIPVisionConfig


class CLIPVisionTower(nn.Module):
    def __init__(self, vision_tower, args, delay_load=False):
        super().__init__()

        self.is_loaded = False

        self.vision_tower_name = vision_tower
        self.select_layer = args.mm_vision_select_layer
        self.select_feature = getattr(args, 'mm_vision_select_feature', 'patch')

        # ##########################################################################
        # print(delay_load)
        # print(getattr(args, 'unfreeze_mm_vision_tower', False))
        # ##########################################################################

        if not delay_load:
            self.load_model()
        elif getattr(args, 'unfreeze_mm_vision_tower', False):
            self.load_model()
        else:
            self.cfg_only = CLIPVisionConfig.from_pretrained(self.vision_tower_name)



    def load_model(self, device_map=None):
        if self.is_loaded:
            print('{} is already loaded, `load_model` called again, skipping.'.format(self.vision_tower_name))
            return

        self.image_processor = CLIPImageProcessor.from_pretrained(self.vision_tower_name)
#        self.vision_tower = CLIPVisionModel.from_pretrained(self.vision_tower_name, device_map=device_map)


        # ##########################################################################
        self.vision_tower = CLIPVisionModel.from_pretrained('laion/CLIP-ViT-bigG-14-laion2B-39B-b160k', device_map=device_map)

        
        SEEC = False
        if SEEC == True:
            print('++++++++++++++++++++++++++ SeeClick Used ++++++++++++++++++++++++++++++')
            print()
            new_state_dict = torch.load('vision_encoder.pth')
            self.vision_tower.load_state_dict(new_state_dict, strict=False)
            print('++++++++++++++++++++++++++ SeeClick Used ++++++++++++++++++++++++++++++')
        else:
            print('++++++++++++++++++++++++++ BigG Used ++++++++++++++++++++++++++++++')
            print()
            print('++++++++++++++++++++++++++ BigG Used ++++++++++++++++++++++++++++++')






        # from transformers import AutoModelForCausalLM, AutoTokenizer
        # model = AutoModelForCausalLM.from_pretrained("/home/kyr/BiaoWu/SeeClick/SeeClick", device_map="cuda", trust_remote_code=True, bf16=True).eval()
        # model.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')  # 设置 device
        # model.transformer.visual.hidden_size = 1664
        # self.vision_tower = model.transformer.visual
        # # self.vision_tower.hidden_size = 1664
        # print(dir(self.vision_tower))

        #import pdb; pdb.set_trace()
        #print(self.vision_tower)
        #print(self.vision_tower.hidden_size)
        print('==========================================================')
        ##########################################################################



        self.vision_tower.requires_grad_(False)
        self.is_loaded = True

    def feature_select(self, image_forward_outs):
        image_features = image_forward_outs.hidden_states[self.select_layer]
        if self.select_feature == 'patch':
            image_features = image_features[:, 1:]
        elif self.select_feature == 'cls_patch':
            image_features = image_features
        else:
            raise ValueError(f'Unexpected select feature: {self.select_feature}')
        return image_features

    @torch.no_grad()
    def forward(self, images):
        if type(images) is list:
            image_features = []
            for image in images:
                image_forward_out = self.vision_tower(image.to(device=self.device, dtype=self.dtype).unsqueeze(0), output_hidden_states=True)
                image_feature = self.feature_select(image_forward_out).to(image.dtype)
                image_features.append(image_feature)
        else:
            image_forward_outs = self.vision_tower(images.to(device=self.device, dtype=self.dtype), output_hidden_states=True)
            image_features = self.feature_select(image_forward_outs).to(images.dtype)

        return image_features

    @property
    def dummy_feature(self):
        return torch.zeros(1, self.hidden_size, device=self.device, dtype=self.dtype)

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

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

    @property
    def config(self):
        if self.is_loaded:
            return self.vision_tower.config
        else:
            return self.cfg_only

    @property
    def hidden_size(self):
        return self.config.hidden_size

    @property
    def num_patches_per_side(self):
        return self.config.image_size // self.config.patch_size

    @property
    def num_patches(self):
        return (self.config.image_size // self.config.patch_size) ** 2