File size: 14,573 Bytes
32b542e |
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 |
# Copyright (c) 2019, AImageLab
import torch
from torch import nn
import torch.nn.functional as F
from torch.autograd import Variable
from uniperceiver.config import configurable
from uniperceiver.functional import expand_tensor
from .decode_strategy import DecodeStrategy
from .build import DECODE_STRATEGY_REGISTRY
from uniperceiver.utils import comm
import math
@DECODE_STRATEGY_REGISTRY.register()
class CaptionBeamSearcherV2(DecodeStrategy):
def data_half(self, data):
if self.fp16:
for k, v in data.items():
if isinstance(v, torch.Tensor) and v.dtype == torch.float32:
data[k] = v.half()
# print(k)
return data
else:
return data
def _select(self, batch_size, beam_size, t, candidate_logprob):
selected_logprob, selected_idx = torch.sort(candidate_logprob.view(batch_size, -1), -1, descending=True)
selected_logprob, selected_idx = selected_logprob[:, :beam_size], selected_idx[:, :beam_size]
return selected_idx, selected_logprob
def _expand_state(self, states, selected_beam, batch_size, beam_size, cur_beam_size):
for i in range(len(states)):
shape = list(states[i].shape)
beam = selected_beam
for _ in shape[1:]:
beam = beam.unsqueeze(-1)
states[i] = torch.gather(states[i].view(*([batch_size, cur_beam_size] + shape[1:])), 1,
beam.expand(*([batch_size, beam_size] + shape[1:])))
states[i] = states[i].view(*([-1, ] + shape[1:]))
def _forward(self, batched_inputs, model):
# only two caption tasks are generative task now!
# for caption tasks, the computations are:
# 1. encode the image sequence; save for further use.
# 2. if no cached encoded dictionary, encode the dictionary and save; otherwise reuse cache.
# 3. compute attention. We use cross attention insted of self attention.
# batched_inputs[kfg.IMAGE] = batched_inputs.pop(kfg.VIDEO).squeeze(1)
inputs = batched_inputs
inputs = self.data_half(inputs)
out_size = batched_inputs.get('OUT_SIZE', 1)
# 0. token embedding
if model.visual_embed is not None:
# ve_out = model.visual_embed(batched_inputs)
# inputs.update(ve_out)
model.visual_embed(inputs)
if model.video_embed is not None:
# ve_out = model.video_embed(batched_inputs)
# inputs.update(ve_out)
model.video_embed(inputs)
if model.token_embed is not None:
# te_out = model.token_embed(batched_inputs)
# inputs.update(te_out)
model.token_embed(inputs)
prompt_data = {}
if model.prompt_embed is not None:
prompt_data = model.prompt_embed(batched_inputs)
prompt_data[kfg.DEEP_PROMPT] = model.prompt and model.deep_prompt
inputs.update(prompt_data)
# 1. encode the image/video sequence.
# bs = inputs[kfg.ATT_FEATS].size(0)
bs = inputs['images'].size(0)
v_input = []
# v_input.append(model._get_sep_embed(inputs, bs))
v_input.append(model._get_sep_embed(inputs.get('spe_token_embed', None), bs))
# v_input.append(inputs[kfg.ATT_FEATS])
# comm._LOCAL_IMAGE_LENGTH = inputs[kfg.ATT_FEATS].shape[1]
comm._LOCAL_IMAGE_LENGTH = inputs['images'].shape[-1]
# add by zjg
if kfg.PROMPT_EMBED in inputs and not model.deep_prompt:
v_input.append(batched_inputs[kfg.PROMPT_EMBED])
v_input = torch.cat(v_input, dim=1)
# ext_u_tmasks = torch.ones((bs, v_input.shape[1], v_input.shape[1]), dtype=v_input.dtype, device=v_input.device)
# ext_u_tmasks = ext_u_tmasks.unsqueeze(1)
# ext_u_tmasks = (1.0 - ext_u_tmasks) * -10000.0
# for img encoder, do not need mask
v_input = {
kfg.MM_EMBEDS: v_input,
# kfg.ATT_FEATS: inputs[kfg.ATT_FEATS],
kfg.TEXT_GEN_MODE: False,
kfg.EXT_U_TOKENS_MASKS: None,
}
# for deep prompt tuning
if prompt_data.get(kfg.DEEP_PROMPT, False):
v_input.update(prompt_data)
# masks = model.get_extended_attention_mask(v_input)
# v_input.update(masks)
# v_input.update( {kfg.EXT_U_TOKENS_MASKS: v_input[kfg.EXT_U_TOKENS_MASKS][:, :, :, 1:]} ) # remove the mask for special token
# vfeats = model.encoder(v_input)[kfg.U_HIDDEN_STATES]
# 2. encode the dictionary - if no pre-computed, add that into input
if getattr(self, 'pre_computed_word_embeds', None) is None:
w_input = []
vocab_size = model.token_embed.embeddings.num_embeddings
w_input.append(model._get_sep_embed(inputs.get('spe_token_embed', None), vocab_size))
# range_slice = torch.arange(start=0, end=vocab_size).unsqueeze(1).to(inputs[kfg.ATT_FEATS].device)
range_slice = torch.arange(start=0, end=vocab_size).unsqueeze(1).to(inputs['images'].device)
# - [HACK] we hardcode the EOT token
eot_to_append = range_slice.new_full(range_slice.shape, 49407)
range_slice_concat_eot = torch.cat([range_slice, eot_to_append], dim=1)
# temp = {
# kfg.U_TOKENS_IDS: range_slice_concat_eot,
# kfg.U_TOKENS_TYPE: torch.zeros_like(range_slice_concat_eot)
# }
temp = {
"shared_targets": [{
"shared_tgt_tokens":range_slice_concat_eot,
},
]
# kfg.U_TOKENS_TYPE: torch.zeros_like(range_slice_concat_eot)
}
# word_embeddings = model.token_embed(temp)['shared_tgt_token_embed']
model.token_embed(temp)
word_embeddings = temp["shared_targets"][0]['shared_tgt_token_embed']
w_input.append(word_embeddings)
w_input = torch.cat(w_input, dim=1)
v_input.update({ kfg.WORD_EMBEDS: w_input })
v_input = self.data_half(v_input)
model.add_tag_embedding(v_input)
enc_out = model.encoder(v_input, return_all=True)
self.pre_computed_word_embeds = enc_out[kfg.WORD_HIDDEN_STATES]
vfeats = enc_out[kfg.U_HIDDEN_STATES]
else:
v_input = self.data_half(v_input)
vfeats = model.encoder(v_input, return_all=True)[kfg.U_HIDDEN_STATES]
# 3. compute attention
comm._CAPTION_GEN_MODE = True
beam_size = self.beam_size
log_probs = []
selected_words = None
seq_logprob = torch.zeros((bs, 1, 1)).cuda() # bs, 1, 1
seq_mask = torch.ones((bs, beam_size, 1)).cuda()
wt = Variable(torch.zeros(bs, dtype=torch.long).cuda().unsqueeze(1)) + self.spe_token_id
u_tokens_type = wt.new_zeros(wt.shape) # [Note] we assume the type tokens are 0.
history_states = vfeats[:-1]
len_prefix = history_states[0].shape[1]
total_history_states = [ history_states[0].new_zeros(beam_size * bs, vfeats[0].shape[1] + self.max_seq_len, vfeats[0].shape[2]) for _ in history_states]
for i, ths in enumerate(total_history_states):
hs = history_states[i]
ths[:hs.shape[0], :hs.shape[1], :] = hs
outputs = []
for t in range(self.max_seq_len):
cur_beam_size = 1 if t == 0 else beam_size
history_states = [ ths[ :(cur_beam_size*bs), :(len_prefix+t), :] for ths in total_history_states]
t_input = {
kfg.U_TOKENS_IDS: wt,
kfg.U_TOKENS_TYPE: u_tokens_type,
kfg.EXT_U_TOKENS_MASKS: None,
kfg.HISTORY_STATES: history_states,
kfg.TIME_STEP: t
}
vt_out = model.token_embed(t_input)
t_input.update(vt_out)
t_input.update({ kfg.MM_EMBEDS: t_input[kfg.U_TOKEN_EMBED] })
if prompt_data.get(kfg.DEEP_PROMPT, False) and prompt_data['PROMPT_EMBED'].shape[1] != t_input[
'MM_EMBEDS'].shape[0]:
prompt_data['PROMPT_EMBED'] = prompt_data[
'PROMPT_EMBED'][:, :1].expand(
-1, t_input['MM_EMBEDS'].shape[0], -1, -1)
t_input.update(prompt_data)
t_input = self.data_half(t_input)
encoder_out = model.encoder(t_input, return_all=True)
pred_input = {
kfg.TEXT_GEN_MODE: True,
kfg.WORD_HIDDEN_STATES: self.pre_computed_word_embeds,
kfg.U_HIDDEN_STATES: encoder_out[kfg.U_HIDDEN_STATES],
kfg.TASK_NAME: batched_inputs[kfg.TASK_NAME]
}
logit = model.predictor(pred_input, force_spe_first=True)[kfg.OUTPUT]
word_logprob = F.log_softmax(logit, dim=-1)
word_logprob = word_logprob.view(bs, cur_beam_size, -1)
candidate_logprob = seq_logprob + word_logprob
# # Mask sequence if it reaches EOS
# if t > 0:
# mask = (selected_words.view(bs, cur_beam_size) != 0).float().unsqueeze(-1) # 为什么是不等于0
# seq_mask = seq_mask * mask
# word_logprob = word_logprob * seq_mask.expand_as(word_logprob)
# old_seq_logprob = seq_logprob.expand_as(candidate_logprob).contiguous()
# old_seq_logprob[:, :, 1:] = -999
# candidate_logprob = seq_mask * candidate_logprob + old_seq_logprob * (1 - seq_mask)
eos_id = 49407
if t > 0:
mask = (selected_words.view(bs, cur_beam_size) != eos_id).float().unsqueeze(-1)
seq_mask = seq_mask * mask
word_logprob = word_logprob * seq_mask.expand_as(word_logprob)
old_seq_logprob = seq_logprob.expand_as(candidate_logprob).contiguous()
old_seq_logprob[:, :, :eos_id] = -999
old_seq_logprob[:, :, eos_id + 1:] = -999
candidate_logprob = seq_mask * candidate_logprob + old_seq_logprob * (1 - seq_mask)
selected_idx, selected_logprob = self._select(bs, beam_size, t, candidate_logprob) # bs beam
selected_beam = torch.div(selected_idx, candidate_logprob.shape[-1], rounding_mode='floor')
selected_words = selected_idx - selected_beam * candidate_logprob.shape[-1]
self._expand_state(history_states, selected_beam, bs, beam_size, cur_beam_size)
seq_logprob = selected_logprob.unsqueeze(-1)
seq_mask = torch.gather(seq_mask, 1, selected_beam.unsqueeze(-1))
outputs = list(torch.gather(o, 1, selected_beam.unsqueeze(-1)) for o in outputs)
outputs.append(selected_words.unsqueeze(-1))
this_word_logprob = torch.gather(word_logprob, 1,
selected_beam.unsqueeze(-1).expand(bs, beam_size, word_logprob.shape[-1]))
this_word_logprob = torch.gather(this_word_logprob, 2, selected_words.unsqueeze(-1))
log_probs = list(
torch.gather(o, 1, selected_beam.unsqueeze(-1).expand(bs, beam_size, 1)) for o in log_probs)
log_probs.append(this_word_logprob)
selected_words = selected_words.view(-1, 1)
# wt = selected_words
if t == 0:
u_tokens_type = expand_tensor(u_tokens_type, beam_size)
wt = expand_tensor(wt, beam_size)
selected_t_input = {
kfg.U_TOKENS_IDS: selected_words,
kfg.U_TOKENS_TYPE: u_tokens_type,
kfg.EXT_U_TOKENS_MASKS: None,
kfg.HISTORY_STATES: history_states,
kfg.TIME_STEP: t
}
selected_vt_out = model.token_embed(selected_t_input)
selected_t_input.update(selected_vt_out)
selected_t_input.update({ kfg.MM_EMBEDS: selected_t_input[kfg.U_TOKEN_EMBED] })
selected_t_prompt_data = dict(prompt_data)
if selected_t_prompt_data.get(kfg.DEEP_PROMPT, False) and selected_t_prompt_data['PROMPT_EMBED'].shape[1] != selected_t_input['MM_EMBEDS'].shape[0]:
selected_t_prompt_data['PROMPT_EMBED'] = selected_t_prompt_data['PROMPT_EMBED'][:, :1].expand(
-1, selected_t_input['MM_EMBEDS'].shape[0], -1, -1)
selected_t_input.update(selected_t_prompt_data)
selected_t_input = self.data_half(selected_t_input)
selected_encoder_out = model.encoder(selected_t_input, return_all=True)
for i, ths in enumerate(total_history_states):
hs = history_states[i]
ths[:hs.shape[0], :hs.shape[1], :] = hs
ths[:hs.shape[0], hs.shape[1], :] = selected_encoder_out[kfg.U_HIDDEN_STATES][i].squeeze(1)
# expand_keys = {
# kfg.ATT_FEATS,
# kfg.GLOBAL_FEATS,
# kfg.ATT_MASKS,
# kfg.EXT_ATT_MASKS,
# kfg.P_ATT_FEATS,
# kfg.EXT_G_TOKENS_MASKS,
# kfg.G_TOKENS_TYPE
# }
# for key in expand_keys:
# if key in inputs:
# if isinstance(inputs[key], list):
# inputs[key] = inputs[key][-1] # usually is ATT_FEATS in TDEN
# tensor = expand_tensor(inputs[key], beam_size)
# inputs.update({ key: tensor })
outputs = torch.cat(outputs, -1)
if self.len_penalty > 0:
step = outputs.ne(49407).sum(-1, keepdim=True) + 1
seq_logprob /= step ** self.len_penalty
seq_logprob, sort_idxs = torch.sort(seq_logprob, 1, descending=True)
outputs = torch.gather(outputs, 1, sort_idxs.expand(bs, beam_size, self.max_seq_len))
log_probs = torch.cat(log_probs, -1)
log_probs = torch.gather(log_probs, 1, sort_idxs.expand(bs, beam_size, self.max_seq_len))
outputs = outputs.contiguous()[:, :out_size]
log_probs = log_probs.contiguous()[:, :out_size]
if out_size == 1:
outputs = outputs.squeeze(1)
log_probs = log_probs.squeeze(1)
comm._CAPTION_GEN_MODE = False
return {
kfg.IDS: batched_inputs[kfg.IDS],
kfg.G_SENTS_IDS: outputs,
kfg.G_LOGP: log_probs
}
|