File size: 11,908 Bytes
fc1c2b8 | 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 | import torch
from torch import nn
import torch.nn.functional as F
from torch.nn import BCEWithLogitsLoss
from transformers import AutoModel, AutoConfig
def mean_pooling(model_output, attention_mask):
token_embeddings = model_output[0] #First element of model_output contains all token embeddings
input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)
class BaseEncoder(nn.Module):
def __init__(self, len_tokenizer, model='huawei-noah/TinyBERT_General_4L_312D'):
super().__init__()
self.transformer = AutoModel.from_pretrained(model)
self.transformer.resize_token_embeddings(len_tokenizer)
def forward(self, input_ids, attention_mask):
output = self.transformer(input_ids, attention_mask)
return output
# self-supervised contrastive model
class ContrastiveSelfSupervisedPretrainModel(nn.Module):
def __init__(self, len_tokenizer, model='huawei-noah/TinyBERT_General_4L_312D', ssv=True, pool=False, proj='mlp', temperature=0.07, num_augments=2):
super().__init__()
self.ssv = ssv
self.pool = pool
self.proj = proj
self.temperature = temperature
self.num_augments = num_augments
self.criterion = SupConLoss(self.temperature)
self.encoder = BaseEncoder(len_tokenizer, model)
self.config = self.encoder.transformer.config
self.contrastive_head = ContrastivePretrainHead(self.config.hidden_size, self.proj)
def forward(self, input_ids, attention_mask, labels):
additional_outputs = []
if self.pool:
output_left = self.encoder(input_ids, attention_mask)
output_left = mean_pooling(output_left, attention_mask)
for num in range(self.num_augments-1):
output_right = self.encoder(input_ids, attention_mask)
output_right = mean_pooling(output_right, attention).unsqueeze(1)
additional_outputs.append(output_right)
else:
output_left = self.encoder(input_ids, attention_mask)['pooler_output'].unsqueeze(1)
for num in range(self.num_augments-1):
additional_outputs.append(self.encoder(input_ids, attention_mask)['pooler_output'].unsqueeze(1))
output = torch.cat((output_left, *additional_outputs), 1)
output = F.normalize(output, dim=-1)
proj_output = self.contrastive_head(output)
proj_output = F.normalize(proj_output, dim=-1)
if self.ssv:
loss = self.criterion(proj_output)
else:
loss = self.criterion(proj_output, labels)
return ((loss,))
# supervised contrastive model
class ContrastivePretrainModel(nn.Module):
def __init__(self, len_tokenizer, model='huawei-noah/TinyBERT_General_4L_312D', pool=True, proj='mlp', temperature=0.07):
super().__init__()
self.pool = pool
self.proj = proj
self.temperature = temperature
self.criterion = SupConLoss(self.temperature)
self.encoder = BaseEncoder(len_tokenizer, model)
self.config = self.encoder.transformer.config
def forward(self, input_ids, attention_mask, labels, input_ids_right, attention_mask_right):
if self.pool:
output_left = self.encoder(input_ids, attention_mask)
output_left = mean_pooling(output_left, attention_mask)
output_right = self.encoder(input_ids_right, attention_mask_right)
output_right = mean_pooling(output_right, attention_mask_right)
else:
output_left = self.encoder(input_ids, attention_mask)['pooler_output']
output_right = self.encoder(input_ids_right, attention_mask_right)['pooler_output']
output = torch.cat((output_left.unsqueeze(1), output_right.unsqueeze(1)), 1)
output = F.normalize(output, dim=-1)
loss = self.criterion(output, labels)
return ((loss,))
class ContrastivePretrainHead(nn.Module):
def __init__(self, hidden_size, proj='mlp'):
super().__init__()
if proj == 'linear':
self.proj = nn.Linear(hidden_size, hidden_size)
elif proj == 'mlp':
self.proj = nn.Sequential(
nn.Linear(hidden_size, hidden_size),
nn.ReLU(),
nn.Linear(hidden_size, hidden_size)
)
def forward(self, hidden_states):
x = self.proj(hidden_states)
return x
# cross-entropy fine-tuning model
class ContrastiveClassifierModel(nn.Module):
def __init__(self, len_tokenizer, checkpoint_path, model='huawei-noah/TinyBERT_General_4L_312D', pool=True, comb_fct='concat-abs-diff-mult', frozen=True, pos_neg=False):
super().__init__()
self.pool = pool
self.frozen = frozen
self.checkpoint_path = checkpoint_path
self.comb_fct = comb_fct
self.pos_neg = pos_neg
self.encoder = BaseEncoder(len_tokenizer, model)
self.config = self.encoder.transformer.config
if self.pos_neg:
self.criterion = BCEWithLogitsLoss(pos_weight=torch.Tensor([pos_neg]))
else:
self.criterion = BCEWithLogitsLoss()
self.classification_head = ClassificationHead(self.config, self.comb_fct)
if self.checkpoint_path:
checkpoint = torch.load(self.checkpoint_path)
self.load_state_dict(checkpoint, strict=False)
if self.frozen:
for param in self.encoder.parameters():
param.requires_grad = False
def forward(self, input_ids, attention_mask, labels, input_ids_right, attention_mask_right):
if self.pool:
output_left = self.encoder(input_ids, attention_mask)
output_left = mean_pooling(output_left, attention_mask)
output_right = self.encoder(input_ids_right, attention_mask_right)
output_right = mean_pooling(output_right, attention_mask_right)
else:
output_left = self.encoder(input_ids, attention_mask)['pooler_output']
output_right = self.encoder(input_ids_right, attention_mask_right)['pooler_output']
if self.comb_fct == 'concat-abs-diff':
output = torch.cat((output_left, output_right, torch.abs(output_left - output_right)), -1)
elif self.comb_fct == 'concat-mult':
output = torch.cat((output_left, output_right, output_left * output_right), -1)
elif self.comb_fct == 'concat':
output = torch.cat((output_left, output_right), -1)
elif self.comb_fct == 'abs-diff':
output = torch.abs(output_left - output_right)
elif self.comb_fct == 'mult':
output = output_left * output_right
elif self.comb_fct == 'abs-diff-mult':
output = torch.cat((torch.abs(output_left - output_right), output_left * output_right), -1)
elif self.comb_fct == 'concat-abs-diff-mult':
output = torch.cat((output_left, output_right, torch.abs(output_left - output_right), output_left * output_right), -1)
proj_output = self.classification_head(output)
loss = self.criterion(proj_output.view(-1), labels.float())
proj_output = torch.sigmoid(proj_output)
return (loss, proj_output)
class ClassificationHead(nn.Module):
def __init__(self, config, comb_fct):
super().__init__()
if comb_fct in ['concat-abs-diff', 'concat-mult']:
self.hidden_size = 3 * config.hidden_size
elif comb_fct in ['concat', 'abs-diff-mult']:
self.hidden_size = 2 * config.hidden_size
elif comb_fct in ['abs-diff', 'mult']:
self.hidden_size = config.hidden_size
elif comb_fct in ['concat-abs-diff-mult']:
self.hidden_size = 4 * config.hidden_size
classifier_dropout = config.hidden_dropout_prob
self.dropout = nn.Dropout(classifier_dropout)
self.out_proj = nn.Linear(self.hidden_size, 1)
def forward(self, features):
x = self.dropout(features)
x = self.out_proj(x)
return x
class SupConLoss(nn.Module):
"""Supervised Contrastive Learning: https://arxiv.org/pdf/2004.11362.pdf.
It also supports the unsupervised contrastive loss in SimCLR"""
def __init__(self, temperature=0.07, contrast_mode='all',
base_temperature=0.07):
super(SupConLoss, self).__init__()
self.temperature = temperature
self.contrast_mode = contrast_mode
self.base_temperature = base_temperature
def forward(self, features, labels=None, mask=None):
"""Compute loss for model. If both `labels` and `mask` are None,
it degenerates to SimCLR unsupervised loss:
https://arxiv.org/pdf/2002.05709.pdf
Args:
features: hidden vector of shape [bsz, n_views, ...].
labels: ground truth of shape [bsz].
mask: contrastive mask of shape [bsz, bsz], mask_{i,j}=1 if sample j
has the same class as sample i. Can be asymmetric.
Returns:
A loss scalar.
"""
device = (torch.device('cuda')
if features.is_cuda
else torch.device('cpu'))
if len(features.shape) < 3:
raise ValueError('`features` needs to be [bsz, n_views, ...],'
'at least 3 dimensions are required')
if len(features.shape) > 3:
features = features.view(features.shape[0], features.shape[1], -1)
batch_size = features.shape[0]
if labels is not None and mask is not None:
raise ValueError('Cannot define both `labels` and `mask`')
elif labels is None and mask is None:
mask = torch.eye(batch_size, dtype=torch.float32).to(device)
elif labels is not None:
labels = labels.contiguous().view(-1, 1)
if labels.shape[0] != batch_size:
raise ValueError('Num of labels does not match num of features')
mask = torch.eq(labels, labels.T).float().to(device)
else:
mask = mask.float().to(device)
contrast_count = features.shape[1]
contrast_feature = torch.cat(torch.unbind(features, dim=1), dim=0)
if self.contrast_mode == 'one':
anchor_feature = features[:, 0]
anchor_count = 1
elif self.contrast_mode == 'all':
anchor_feature = contrast_feature
anchor_count = contrast_count
else:
raise ValueError('Unknown mode: {}'.format(self.contrast_mode))
# compute logits
anchor_dot_contrast = torch.div(
torch.matmul(anchor_feature, contrast_feature.T),
self.temperature)
# for numerical stability
logits_max, _ = torch.max(anchor_dot_contrast, dim=1, keepdim=True)
logits = anchor_dot_contrast - logits_max.detach()
# tile mask
mask = mask.repeat(anchor_count, contrast_count)
# mask-out self-contrast cases
logits_mask = torch.scatter(
torch.ones_like(mask),
1,
torch.arange(batch_size * anchor_count).view(-1, 1).to(device),
0
)
mask = mask * logits_mask
# compute log_prob
exp_logits = torch.exp(logits) * logits_mask
log_prob = logits - torch.log(exp_logits.sum(1, keepdim=True))
# compute mean of log-likelihood over positive
mean_log_prob_pos = (mask * log_prob).sum(1) / mask.sum(1)
# loss
loss = - (self.temperature / self.base_temperature) * mean_log_prob_pos
loss = loss.view(anchor_count, batch_size).mean()
return loss |