File size: 11,764 Bytes
14bd491
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
from llava.model.multimodal_encoder.processor import Blip2ImageTrainProcessor
from transformers import AutoTokenizer
from llava.constants import MM_TOKEN_INDEX, DEFAULT_IMAGE_PATCH_TOKEN, DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN, \
    DEFAULT_VIDEO_PATCH_TOKEN, DEFAULT_VIDEO_TOKEN, DEFAULT_VIDEO_START_TOKEN, DEFAULT_VIDEO_END_TOKEN
from llava.conversation import conv_templates
from llava.model.multimodal_encoder.qformer import BertConfig, BertLMHeadModel, BertModel
from llava.model.multimodal_projector.builder import build_vision_projector
from llava.model.utils import LayerNorm
from llava.model.multimodal_encoder.eva_clip_encoder import EvaClipVisionTower
import torch
from llava.mm_utils import tokenizer_image_token, process_images_v2, KeywordsStoppingCriteria
import numpy as np
from PIL import Image
import os
import torch.nn as nn
from transformers import AutoConfig
from collections import OrderedDict

import torch_neuronx

NUM_SEGMENTS = 10

def generate_input_ids(tokenizer):
    conv = conv_templates['thoth'].copy()
    qs = "Describe the following video in detail."
    qs = DEFAULT_VIDEO_START_TOKEN + DEFAULT_VIDEO_TOKEN + DEFAULT_VIDEO_END_TOKEN + '\n' + qs
    conv.append_message(conv.roles[0], qs)
    conv.append_message(conv.roles[1], None)
    prompt = conv.get_prompt()
    input_ids = tokenizer_image_token(prompt, tokenizer, MM_TOKEN_INDEX, return_tensors='pt').unsqueeze(0)
    return input_ids, conv


def generate_images(frame_folder, image_processor, model_cfg):
    images = load_frames(frame_folder)
    if len(images) > NUM_SEGMENTS:
        images = uniform_sample(images, NUM_SEGMENTS)
    return process_images_v2(images, image_processor, model_cfg)


def uniform_sample(frames, num_segments):
    indices = np.linspace(start=0, stop=len(frames) - 1, num=num_segments).astype(int)
    frames = [frames[ind] for ind in indices]
    return frames


def load_frames(frames_dir):
    results = []
    image_files = [(int(os.path.splitext(img)[0]), img) for img in os.listdir(frames_dir) if img.endswith('jpg')]
    image_files = sorted(image_files, key=lambda img: img[0])

    for frame_name in image_files:
        image_path = f"{frames_dir}/{frame_name[1]}"
        image = Image.open(image_path).convert('RGB')
        results.append(image)
    return results


class MASPVision(torch.nn.Module):

    def __init__(self, config):
        super().__init__()
        # device = 'cuda:0'
        device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
        device_map = {"": 0}
        config.vit_model_path = 'eva_vit_g.pth'
        vision_tower = EvaClipVisionTower("eva-vit-g", config, delay_load=True)
        vision_tower.load_model(device_map=device_map)
        vision_tower.to(device=device, dtype=torch.float16)

        image_processor = Blip2ImageTrainProcessor(
            image_size=config.img_size,
            is_training=False)

        cross_attention_freq = 2
        vision_width = vision_tower.hidden_size
        num_query_token = config.num_query_token
        ln_vision = LayerNorm(vision_width)
        encoder_config = BertConfig.from_pretrained("bert-base-uncased")
        encoder_config.encoder_width = vision_width
        # insert cross-attention layer every other block
        encoder_config.add_cross_attention = True
        encoder_config.cross_attention_freq = cross_attention_freq
        encoder_config.query_length = num_query_token
        # Qformer = BertLMHeadModel(config=encoder_config)
        self.bert = BertModel(encoder_config, add_pooling_layer=False)
        self.bert.embeddings.word_embeddings = None
        self.bert.embeddings.position_embeddings = None

        for layer in self.bert.encoder.layer:
            layer.output = None
            layer.intermediate = None

        query_tokens = nn.Parameter(
            torch.zeros(1, num_query_token, encoder_config.hidden_size)
        )
        query_tokens.data.normal_(mean=0.0, std=encoder_config.initializer_range)

        frame_position_encoding = nn.Embedding(
            config.max_num_segments,
            encoder_config.hidden_size
        )

        mm_projector = build_vision_projector(config)

        self.vision_tower = vision_tower
        # self.qformer = Qformer
        self.projector = mm_projector
        self.query_tokens = query_tokens
        self.ln_vision = ln_vision
        self.frame_position_encoding = frame_position_encoding

    def forward(self, images):
        # images: [num_frames, patches, 3, image_size, image_size]
        image_features = self.vision_tower(images.flatten(0, 1))
        image_features = self.ln_vision(image_features)
        attn_mask = torch.ones(image_features.size()[:-1], dtype=torch.long).to(
            image_features.device)  # [num_frames * num_patches, 256]
        query_tokens = self.query_tokens.expand(image_features.shape[0], -1, -1)  # [num_frames * num_patches, 32, 768]
        dtype_ = self.vision_tower.dtype
        image_features = self.bert(
            query_embeds=query_tokens.to(dtype_),
            encoder_hidden_states=image_features.to(dtype_),
            encoder_attention_mask=attn_mask,
            return_dict=True
        ).last_hidden_state.to(dtype_)
        frame_ids = torch.arange(images.shape[0], dtype=torch.long, device=image_features.device).unsqueeze(1)
        frame_ids = frame_ids.repeat(1, images.shape[1]).flatten(0, 1)  # [num_frames * num_patches]
        image_features += self.frame_position_encoding(frame_ids).unsqueeze(-2)  # [num_frames, 1, 768]
        return self.projector(image_features)

        # zheng add

    def forward_features(self, a, b):
        # images: [num_frames, patches, 3, image_size, image_size]
        images = a
        image_features = b
        image_features = self.ln_vision(image_features)
        attn_mask = torch.ones(image_features.size()[:-1], dtype=torch.long).to(
            image_features.device)  # [num_frames * num_patches, 256]
        query_tokens = self.query_tokens.expand(image_features.shape[0], -1, -1)  # [num_frames * num_patches, 32, 768]
        dtype_ = self.vision_tower.dtype
        image_features = self.bert(
            query_embeds=query_tokens.to(dtype_),
            encoder_hidden_states=image_features.to(dtype_),
            encoder_attention_mask=attn_mask,
            return_dict=True
        ).last_hidden_state.to(dtype_)
        frame_ids = torch.arange(images.shape[0], dtype=torch.long, device=image_features.device).unsqueeze(1)
        frame_ids = frame_ids.repeat(1, images.shape[1]).flatten(0, 1)  # [num_frames * num_patches]
        image_features += self.frame_position_encoding(frame_ids).unsqueeze(-2)  # [num_frames, 1, 768]
        return self.projector(image_features)


if __name__ == '__main__':
    frame_folder = './v12044gd0000cl5c6rfog65i2eoqcqig'
    tokenizer_dir = '../tokenizer_dir'
    # device = 'cuda:0'
    device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

    tokenizer = AutoTokenizer.from_pretrained(tokenizer_dir)
    config = AutoConfig.from_pretrained(tokenizer_dir, trust_remote_code=True)
    tokenizer.add_tokens(
        [DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN, DEFAULT_VIDEO_START_TOKEN, DEFAULT_VIDEO_END_TOKEN],
        special_tokens=True)

    image_processor = Blip2ImageTrainProcessor(
        image_size=config.img_size,
        is_training=False)

    input_ids, conv = generate_input_ids(tokenizer)

    # images = generate_images(frame_folder, image_processor, config).to(device).half() # [num_frames, patches, 3, image_size, image_size]

    # zheng
    images = generate_images(frame_folder, image_processor, config).to(device)

    vision_module = MASPVision(config=config)

    input_ids = input_ids[0].to(device)  # [token_len]
    # new_vision_state_dict = torch.load('new_vision_state_dict.pth')
    new_vision_state_dict = torch.load('new_vision_state_dict.pth', map_location=device)

    # vision_state_dict = torch.load('masp_vision_statedict.pth', map_location="cuda:0")
    # new_vision_state_dict = OrderedDict()
    # for k, v in vision_state_dict.items():
    #     if 'qformer' in k:
    #         new_key = k[8:]
    #         new_vision_state_dict[new_key] = v
    #     else:
    #         new_vision_state_dict[k] = v

    vision_module.load_state_dict(new_vision_state_dict)
    vision_module = vision_module.eval()
    vision_module = vision_module.to(device)

    # vision_module.to(torch.float16)
    # zheng add
    vision_module.to(torch.float32)

    # zheng add
    vision_module_neuron = torch.jit.load("./neuron_eva_vit_base.pt")
    vision_module_neuron = vision_module_neuron.eval()
    # output=vision_module_neuron(images)

    padding_idx = config.pad_token_id
    embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx)
    embed_weight = torch.load('embed_tokens.pth')
    embed_tokens.load_state_dict(embed_weight)
    embed_tokens = embed_tokens.eval()
    embed_tokens.to(torch.float16).to(device)
    # vision_module = vision_module.eval()
    # vision_state_dict = vision_module.state_dict()
    # torch.save(vision_state_dict, 'masp_vision_statedict.pth')

    # infernece
    # print(images.shape) # [10, 7, 3, 224, 224]
    # dummy_images = torch.rand(10, 7, 3, 224, 224).to(model.device)
    # scripted_vision_module = torch.jit.script(vision_module)
    # print('begin to trace')
    # traced_vision_module = torch.jit.trace(vision_module, (images))
    # traced_vision_module.save('traced_vision_module.pt')
    # loaded = torch.jit.load('traced_vision_module.pt')

    import time

    start = time.time()
    with torch.inference_mode():
        # get image feature

        # image_features = vision_module(images).flatten(0, 1) # [num_frames * num_patches * num_query_token, 4096]

        image_features = torch.Tensor()  # init a tensor
        for image in images:
            output = vision_module_neuron(image)
            output = output[:, 1:].to(torch.float32)
            if len(image_features) == 0:
                image_features = output
            else:
                image_features = torch.cat([image_features, output], dim=0)

        # zheng [70, 256, 1408]
        image_features = vision_module.forward_features(images, image_features)
        image_features = image_features.flatten(0, 1)
        print(image_features.shape)  # zheng [70, 32, 4096]

        image_features.to(device=device, dtype=torch.float16)
        image_features_numpy = image_features.detach().cpu().numpy()
        # image_features_saved = np.load('image_features_numpy.npy')
        # print(np.sum(image_features_numpy -image_features_saved ))
        # image_features_numpy = image_features.detach().cpu().numpy
        # np.save('image_features_numpy.npy', image_features_numpy)
        # print('images features shape', image_features.shape)
        # image_features = loaded(images).flatten(0, 1)
        # concat with text features
        vision_token_indice = torch.where(input_ids == MM_TOKEN_INDEX)[0][0]
        pre_text_token = embed_tokens(input_ids[:vision_token_indice])  # zheng [32, 4096]
        post_text_token = embed_tokens(input_ids[vision_token_indice + 1:])

        inputs_embeds = torch.cat([pre_text_token, image_features, post_text_token]).unsqueeze(
            0)  # [1, num_token, 4096]

        print("Inference time:", time.time() - start)

        input_embeds_numpy = inputs_embeds.detach().cpu().numpy()
        image_embeds_saved = np.load('inputs_embeds.npy')
        diff = np.sum(input_embeds_numpy - image_embeds_saved)
        print('diff with saved in the disk', diff)
        # print('inputs embeds numpy shape', input_embeds_numpy.shape)
        # np.save('inputs_embeds.npy', input_embeds_numpy)