File size: 16,417 Bytes
7cdf421 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 |
import random
import torch
from torch.nn.utils import rnn
import io
import json
import logging
import os
import pickle
import re
import shutil
import urllib
import urllib.error
import urllib.request
from typing import Optional
from urllib.parse import urlparse
def truncate_caption(caption: str) -> str:
"""Truncate captions at periods and newlines."""
caption = caption.strip('\n')
trunc_index = caption.find('\n') + 1
if trunc_index <= 0:
trunc_index = caption.find('.') + 1
if trunc_index > 0:
caption = caption[:trunc_index]
return caption
def build_one_instance_for_pgpt4(tokenizer, conversation):
text_list = []
turn_num = len(conversation)
input_ids, target_ids = [], []
for i in range(turn_num):
turn = conversation[i]
role = turn['from']
if i == 0: # the first human turn
assert role == 'human'
text = '### Human: </Img> ' + turn['value'] + '\n### Assistant: '
one_input_id = tokenizer(text, add_special_tokens=False).input_ids
input_ids += one_input_id
target_ids += [-100] * len(one_input_id) # do not perform loss regression on human prompt
else:
if role == 'human':
text = 'Human: ' + turn['value'] + '\n### Assistant: '
one_input_id = tokenizer(text, add_special_tokens=False).input_ids
input_ids += one_input_id
target_ids += [-100] * len(one_input_id)
elif role == 'gpt':
text = turn['value'] + '\n###'
one_input_id = tokenizer(text, add_special_tokens=False).input_ids
input_ids += one_input_id
target_ids += one_input_id
else:
raise Exception('Wrong Role!!!')
text_list.append(text)
assert len(input_ids) == len(target_ids)
return text_list, input_ids, target_ids
def build_one_instance_for_cc3m(tokenizer, conversation):
text_list = []
input_ids, target_ids = [], []
turn_num = len(conversation)
for i in range(turn_num):
turn = conversation[i]
role = turn['from']
if i == 0: # the first human turn
assert role == 'human'
text = '### Human: ' + turn['value'] + '\n### Assistant: '
one_input_id = tokenizer(text, add_special_tokens=False).input_ids
input_ids += one_input_id
target_ids += [-100] * len(one_input_id) # do not perform loss regression on human prompt
else:
if role == 'human':
text = 'Human: ' + turn['value'] + '\n### Assistant: '
one_input_id = tokenizer(text, add_special_tokens=False).input_ids
input_ids += one_input_id
target_ids += [-100] * len(one_input_id)
elif role == 'gpt':
if 'image_name' in turn.keys():
img_tokens = ' '.join([f'[IMG{i}]' for i in range(8)])
text = turn['value'] + ' ' + img_tokens + '\n###'
else:
text = turn['value'] + '\n###'
one_input_id = tokenizer(text, add_special_tokens=False).input_ids
input_ids += one_input_id
target_ids += one_input_id
# if 'image_name' in turn.keys():
# img_tokens = ' '.join([f'[IMG{i}]' for i in range(8)])
# img_input_ids = tokenizer(img_tokens, add_special_tokens=False).input_ids
# input_ids += img_input_ids
# target_ids += img_input_ids
else:
raise Exception('Wrong Role!!!')
text_list.append(text)
assert len(input_ids) == len(target_ids)
return text_list, input_ids, target_ids
def build_one_instance_for_cc3m_1(tokenizer, conversation, num_img_tokens=8):
text_list = []
input_ids, target_ids = [], []
turn_num = len(conversation)
for i in range(turn_num):
turn = conversation[i]
role = turn['from']
if i == 0: # the first human turn
assert role == 'human'
text = turn['value'] + '\n### Assistant: '
# text = turn['value']
one_input_id = tokenizer(text, add_special_tokens=False).input_ids
input_ids += one_input_id
target_ids += one_input_id # do not perform loss regression on human prompt
else:
if role == 'human':
text = turn['value']
one_input_id = tokenizer(text, add_special_tokens=False).input_ids
input_ids += one_input_id
target_ids += one_input_id
elif role == 'gpt':
# if 'image_name' in turn.keys():
# img_tokens = ' '.join([f'[IMG{i}]' for i in range(num_img_tokens)])
# text = turn['value'] + img_tokens
# else:
# text = turn['value']
text = ' '.join([f'[IMG{i}]' for i in range(num_img_tokens)])
one_input_id = tokenizer(text, add_special_tokens=False).input_ids
input_ids += one_input_id
target_ids += one_input_id
# if 'image_name' in turn.keys():
# img_tokens = ' '.join([f'[IMG{i}]' for i in range(8)])
# img_input_ids = tokenizer(img_tokens, add_special_tokens=False).input_ids
# input_ids += img_input_ids
# target_ids += img_input_ids
else:
raise Exception('Wrong Role!!!')
text_list.append(text)
assert len(input_ids) == len(target_ids)
return text_list, input_ids, target_ids
def build_one_instance_for_webvid(tokenizer, conversation, num_video_tokens=8):
text_list = []
input_ids, target_ids = [], []
# text = '### Human: ' + conversation + '\n### Assistant: '
# one_input_id = tokenizer(text, add_special_tokens=False).input_ids
# input_ids += one_input_id
# target_ids += one_input_id # do not perform loss regression on human prompt
video_tokens = ' '.join([f'[VID{i}]' for i in range(num_video_tokens)])
text = conversation + video_tokens
text_list.append(text)
one_input_id = tokenizer(text, add_special_tokens=False).input_ids
input_ids += one_input_id
target_ids += one_input_id
assert len(input_ids) == len(target_ids)
return text_list, input_ids, target_ids
def process_batch_instance(tokenizer, batch_of_conversations, max_tgt_len, dataset='cc3m',
num_img_tokens=8, num_video_tokens=8):
batch_input_ids, batch_target_ids = [], []
for conversation in batch_of_conversations:
if dataset == "pgpt4":
_, one_input_ids, one_target_ids = build_one_instance_for_pgpt4(tokenizer, conversation)
elif dataset == 'cc3m' or dataset == 'coco2017':
_, one_input_ids, one_target_ids = build_one_instance_for_cc3m_1(tokenizer, conversation, num_img_tokens)
elif dataset == 'webvid':
_, one_input_ids, one_target_ids = build_one_instance_for_webvid(tokenizer, conversation, num_video_tokens)
else:
raise Exception("not support dataset name, it should be pgpt4 or cc3m")
batch_input_ids.append(torch.LongTensor(one_input_ids))
batch_target_ids.append(torch.LongTensor(one_target_ids))
input_ids = rnn.pad_sequence(batch_input_ids, batch_first=True, padding_value=tokenizer.pad_token_id)
target_ids = rnn.pad_sequence(batch_target_ids, batch_first=True, padding_value=-100)
assert input_ids.size() == target_ids.size()
input_ids = input_ids[:, :max_tgt_len]
target_ids = target_ids[:, :max_tgt_len]
attention_mask = input_ids.ne(tokenizer.pad_token_id)
assert attention_mask.size() == input_ids.size()
return input_ids, target_ids, attention_mask.long()
def mask_token(inputs, tokenizer, mlm_probability, vocab_size=None, special_tokens_mask=None):
"""
randomly mask some input tokens
"""
indices_replaced = torch.bernoulli(torch.full(inputs.shape, mlm_probability)).bool()
inputs[indices_replaced] = tokenizer.convert_tokens_to_ids(tokenizer.mask_token)
return inputs
def build_one_instance_stage_1(tokenizer, captions, prompt=''):
input_ids, target_ids = [], []
texts = ''
text = '</Img> ' + prompt + '\n### Assistant: '
texts += text
one_input_id = tokenizer(text, add_special_tokens=False).input_ids
input_ids += one_input_id
target_ids += [-100] * len(one_input_id) # do not perform loss regression on human prompt
text = captions + '\n###'
texts += text
one_input_id = tokenizer(text, add_special_tokens=False).input_ids
input_ids += one_input_id
target_ids += one_input_id
return input_ids, target_ids
def process_batch_stage_1(tokenizer, batch_of_captions, max_tgt_len, prompt=''):
batch_input_ids, batch_target_ids = [], []
for caption in batch_of_captions:
one_input_ids, one_target_ids = build_one_instance_stage_1(tokenizer, caption, prompt)
batch_input_ids.append(torch.LongTensor(one_input_ids))
batch_target_ids.append(torch.LongTensor(one_target_ids))
input_ids = rnn.pad_sequence(batch_input_ids, batch_first=True, padding_value=tokenizer.pad_token_id)
target_ids = rnn.pad_sequence(batch_target_ids, batch_first=True, padding_value=-100)
assert input_ids.size() == target_ids.size()
input_ids = input_ids[:, :max_tgt_len]
target_ids = target_ids[:, :max_tgt_len]
attention_mask = input_ids.ne(tokenizer.pad_token_id)
assert attention_mask.size() == input_ids.size()
return input_ids, target_ids, attention_mask.long()
def build_one_instance_stage_2(tokenizer, captions, num_signal_tokens=4, MODALITY='image'):
input_ids, target_ids = [], []
text = captions + '\n### Assistant: '
one_input_id = tokenizer(text, add_special_tokens=False).input_ids
input_ids += one_input_id
target_ids += [-100] * len(one_input_id) # do not perform loss regression on human prompt
if MODALITY == 'image':
signal_tokens = ' '.join([f'[IMG{i}]' for i in range(num_signal_tokens)])
elif MODALITY == 'video':
signal_tokens = ' '.join([f'[VID{i}]' for i in range(num_signal_tokens)])
elif MODALITY == 'audio':
signal_tokens = ' '.join([f'[AUD{i}]' for i in range(num_signal_tokens)])
else:
signal_tokens = ''
text = captions + signal_tokens + '\n###'
one_input_id = tokenizer(text, add_special_tokens=False).input_ids
input_ids += one_input_id
target_ids += one_input_id
return input_ids, target_ids
def process_batch_stage_2(tokenizer, batch_of_captions, max_tgt_len, num_signal_tokens=4, MODALITY='image'):
"""
:param mode: the target modality
:param num_tokens: the number of generated signal tokens for generation
"""
batch_input_ids, batch_target_ids = [], []
# batch_caption_lists = []
for captions in batch_of_captions:
one_input_ids, one_target_ids = build_one_instance_stage_2(tokenizer, captions,
num_signal_tokens=num_signal_tokens,
MODALITY=MODALITY)
batch_input_ids.append(torch.LongTensor(one_input_ids))
batch_target_ids.append(torch.LongTensor(one_target_ids))
# batch_caption_lists.append(caption)
input_ids = rnn.pad_sequence(batch_input_ids, batch_first=True, padding_value=tokenizer.pad_token_id)
target_ids = rnn.pad_sequence(batch_target_ids, batch_first=True, padding_value=-100)
assert input_ids.size() == target_ids.size()
input_ids = input_ids[:, :max_tgt_len]
target_ids = target_ids[:, :max_tgt_len]
attention_mask = input_ids.ne(tokenizer.pad_token_id)
assert attention_mask.size() == input_ids.size()
return input_ids, target_ids, attention_mask.long()
# def process_batch_stage_2(tokenizer, batch_of_captions, )
def build_one_instance_stage_3(tokenizer, conversation, img_tokens=4, vid_tokens=24, aud_tokens=8):
text_list = []
turn_num = len(conversation)
input_ids, target_ids = [], []
for i in range(turn_num):
turn = conversation[i]
role = turn['from']
if i == 0: # the first human turn
assert role == 'human'
if turn['input_modality'] != 'text':
text = '</Img> ' + turn['value'] + '\n### Assistant: '
else:
text = turn['value'] + '\n### Assistant: '
one_input_id = tokenizer(text, add_special_tokens=False).input_ids
input_ids += one_input_id
target_ids += [-100] * len(one_input_id) # do not perform loss regression on human prompt
else:
if role == 'human':
text = 'Human: ' + turn['value'] + '\n### Assistant: '
one_input_id = tokenizer(text, add_special_tokens=False).input_ids
input_ids += one_input_id
target_ids += [-100] * len(one_input_id)
elif role == 'gpt':
if turn['output_modality'] == 'image':
signal_tokens = ' '.join([f'[IMG{i}]' for i in range(img_tokens)])
elif turn['output_modality'] == 'video':
signal_tokens = ' '.join([f'[VID{i}]' for i in range(vid_tokens)])
elif turn['output_modality'] == 'audio':
signal_tokens = ' '.join([f'[AUD{i}]' for i in range(aud_tokens)])
else:
signal_tokens = ''
caption = turn['caption']
text = turn['value'] + signal_tokens + '\n###'
one_input_id = tokenizer(text, add_special_tokens=False).input_ids
input_ids += one_input_id
target_ids += one_input_id
else:
raise Exception('Wrong Role!!!')
text_list.append(text)
assert len(input_ids) == len(target_ids)
return text_list, input_ids, target_ids, caption
def process_batch_stage_3(tokenizer, batch_of_conversations, max_tgt_len, img_tokens=4, vid_tokens=24, aud_tokens=8):
"""
:param mode: the target modality
:param num_tokens: the number of generated signal tokens for generation
"""
batch_input_ids, batch_target_ids = [], []
# batch_caption_lists = []
for conversation in batch_of_conversations:
_, one_input_ids, one_target_ids, caption = build_one_instance_stage_3(tokenizer, conversation,
img_tokens=img_tokens,
vid_tokens=vid_tokens,
aud_tokens=aud_tokens)
batch_input_ids.append(torch.LongTensor(one_input_ids))
batch_target_ids.append(torch.LongTensor(one_target_ids))
# batch_caption_lists.append(caption)
input_ids = rnn.pad_sequence(batch_input_ids, batch_first=True, padding_value=tokenizer.pad_token_id)
target_ids = rnn.pad_sequence(batch_target_ids, batch_first=True, padding_value=-100)
assert input_ids.size() == target_ids.size()
input_ids = input_ids[:, :max_tgt_len]
# if is_mask_token:
# input_ids = mask_token(input_ids, tokenizer, 0.5)
target_ids = target_ids[:, :max_tgt_len]
attention_mask = input_ids.ne(tokenizer.pad_token_id)
assert attention_mask.size() == input_ids.size()
return input_ids, target_ids, attention_mask.long()
def is_url(url_or_filename):
parsed = urlparse(url_or_filename)
return parsed.scheme in ("http", "https")
def l2_loss(u: torch.Tensor, v: torch.Tensor) -> torch.Tensor:
"""
Args:
u: (N, T_I_V_A.txt, D) tensor.
v: (N, T_I_V_A.txt, D) tensor.
Returns:
l1_loss: (N,) tensor of summed L1 loss.
"""
assert u.shape == v.shape, (u.shape, v.shape)
return ((u - v) ** 2).sum(dim=-1) ** 0.5
def get_modality(path_list):
_postfix = os.path.splitext(path_list[0])[-1]
if _postfix == '.jpg':
return 'image'
elif _postfix == '.wav':
return 'audio'
elif _postfix == '.mp4':
return 'video'
else:
raise NotImplementedError
|