File size: 10,134 Bytes
975624b |
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 |
import lightning.pytorch as pl
from transformers import (
AdamW,
AutoModel,
AutoConfig,
get_linear_schedule_with_warmup,
)
from transformers.models.bert.modeling_bert import BertLMPredictionHead
import torch
from torch import nn
from loss import PCCL
import config
class CL_model(pl.LightningModule):
def __init__(
self, n_batches=None, n_epochs=None, lr=None, mlm_weight=None, **kwargs
):
super().__init__()
## Params
self.n_batches = n_batches
self.n_epochs = n_epochs
self.lr = lr
self.mlm_weight = mlm_weight
self.config = AutoConfig.from_pretrained("emilyalsentzer/Bio_ClinicalBERT")
## Encoder
self.bert = AutoModel.from_pretrained(
"emilyalsentzer/Bio_ClinicalBERT", return_dict=True
)
# Unfreeze layers
self.bert_layer_num = sum(1 for _ in self.bert.named_parameters())
self.num_unfreeze_layer = self.bert_layer_num
self.ratio_unfreeze_layer = 0.0
if kwargs:
for key, value in kwargs.items():
if key == "unfreeze" and isinstance(value, float):
assert (
value >= 0.0 and value <= 1.0
), "ValueError: value must be a ratio between 0.0 and 1.0"
self.ratio_unfreeze_layer = value
if self.ratio_unfreeze_layer > 0.0:
self.num_unfreeze_layer = int(
self.bert_layer_num * self.ratio_unfreeze_layer
)
for param in list(self.bert.parameters())[: -self.num_unfreeze_layer]:
param.requires_grad = False
self.lm_head = BertLMPredictionHead(self.config)
# self.projector = nn.Linear(self.bert.config.hidden_size, 128)
print("Model Initialized!")
## Losses
self.cl_loss = PCCL()
self.mlm_loss = nn.CrossEntropyLoss()
## Logs
self.num_batches = 0
self.train_loss, self.val_loss = 0, 0
self.train_loss_cl, self.val_loss_cl = 0, 0
self.train_loss_mlm, self.val_loss_mlm = 0, 0
self.training_step_outputs, self.validation_step_outputs = [], []
def forward(self, input_ids, attention_mask, masked_indices, eval=False):
embs = self.bert(input_ids=input_ids, attention_mask=attention_mask)
cls_tokens = embs.pooler_output
mask_tokens = []
for idx, value in enumerate(masked_indices):
masks = embs.last_hidden_state[idx][value]
avg_mask = torch.mean(masks, dim=0)
mask_tokens.append(avg_mask)
mask_tokens = torch.stack(mask_tokens)
cls_concat_mask = torch.cat((cls_tokens, mask_tokens), dim=1)
if eval is True:
return cls_tokens, mask_tokens, cls_concat_mask
mlm_pred = self.lm_head(embs.last_hidden_state)
mlm_pred = mlm_pred.view(-1, self.config.vocab_size)
return cls_concat_mask, mlm_pred
def training_step(self, batch, batch_idx):
input_ids = batch["input_ids"]
attention_mask = batch["attention_mask"]
mlm_labels = batch["mlm_labels"]
masked_indices = batch["masked_indices"]
tags = batch["tags"]
scores = batch["scores"]
cls_concat_mask, mlm_pred = self(input_ids, attention_mask, masked_indices)
loss_cl = self.cl_loss(cls_concat_mask, tags, scores)
loss_mlm = self.mlm_loss(mlm_pred, mlm_labels.reshape(-1))
loss = (1 - self.mlm_weight) * loss_cl + self.mlm_weight * loss_mlm
logs = {"loss": loss, "loss_cl": loss_cl, "loss_mlm": loss_mlm}
self.training_step_outputs.append(logs)
self.log("train_loss", loss, prog_bar=True, logger=True, sync_dist=True)
self.num_batches += 1
self.train_loss_cl += loss_cl
self.train_loss_mlm += loss_mlm
self.train_loss += loss
if self.num_batches % config.log_every_n_steps == 0:
avg_loss_cl = self.train_loss_cl / self.num_batches
avg_loss_mlm = self.train_loss_mlm / self.num_batches
avg_loss = self.train_loss / self.num_batches
self.log(
"train_avg_cl_loss",
avg_loss_cl,
prog_bar=True,
logger=True,
sync_dist=True,
)
self.log(
"train_avg_mlm_loss",
avg_loss_mlm,
prog_bar=True,
logger=True,
sync_dist=True,
)
self.log(
"train_avg_loss", avg_loss, prog_bar=True, logger=True, sync_dist=True
)
self.train_loss_cl = 0
self.train_loss_mlm = 0
self.train_loss = 0
self.num_batches = 0
return loss
def on_train_epoch_end(self):
e_t_avg_loss = (
torch.stack([x["loss"] for x in self.training_step_outputs])
.mean()
.detach()
.cpu()
.numpy()
)
self.log(
"avg_loss_train_epoch",
e_t_avg_loss.item(),
on_step=False,
on_epoch=True,
sync_dist=True,
)
e_t_avg_loss_cl = (
torch.stack([x["loss_cl"] for x in self.training_step_outputs])
.mean()
.detach()
.cpu()
.numpy()
)
self.log(
"avg_loss_cl_train_epoch",
e_t_avg_loss_cl.item(),
on_step=False,
on_epoch=True,
sync_dist=True,
)
e_t_avg_loss_mlm = (
torch.stack([x["loss_mlm"] for x in self.training_step_outputs])
.mean()
.detach()
.cpu()
.numpy()
)
self.log(
"avg_loss_mlm_train_epoch",
e_t_avg_loss_mlm.item(),
on_step=False,
on_epoch=True,
sync_dist=True,
)
print(
"train_epoch:",
self.current_epoch,
"avg_loss:",
e_t_avg_loss,
"avg_cl_loss:",
e_t_avg_loss_cl,
"avg_mlm_loss:",
e_t_avg_loss_mlm,
)
self.training_step_outputs.clear()
def validation_step(self, batch, batch_idx):
input_ids = batch["input_ids"]
attention_mask = batch["attention_mask"]
mlm_labels = batch["mlm_labels"]
masked_indices = batch["masked_indices"]
tags = batch["tags"]
scores = batch["scores"]
cls_concat_mask, mlm_pred = self(input_ids, attention_mask, masked_indices)
loss_cl = self.cl_loss(cls_concat_mask, tags, scores)
loss_mlm = self.mlm_loss(mlm_pred, mlm_labels.reshape(-1))
loss = (1 - self.mlm_weight) * loss_cl + self.mlm_weight * loss_mlm
logs = {"loss": loss, "loss_cl": loss_cl, "loss_mlm": loss_mlm}
self.validation_step_outputs.append(logs)
self.log("val_loss", loss, prog_bar=True, logger=True, sync_dist=True)
self.num_batches += 1
self.val_loss_cl += loss_cl
self.val_loss_mlm += loss_mlm
self.val_loss += loss
if self.num_batches % config.log_every_n_steps == 0:
avg_loss_cl = self.val_loss_cl / self.num_batches
avg_loss_mlm = self.val_loss_mlm / self.num_batches
avg_loss = self.val_loss / self.num_batches
self.log(
"val_avg_cl_loss",
avg_loss_cl,
prog_bar=True,
logger=True,
sync_dist=True,
)
self.log(
"val_avg_mlm_loss",
avg_loss_mlm,
prog_bar=True,
logger=True,
sync_dist=True,
)
self.log(
"val_avg_loss",
avg_loss,
prog_bar=True,
logger=True,
sync_dist=True,
)
self.val_loss_cl = 0
self.val_loss_mlm = 0
self.val_loss = 0
self.num_batches = 0
return loss
def on_validation_epoch_end(self):
e_v_avg_loss = (
torch.stack([x["loss"] for x in self.validation_step_outputs])
.mean()
.detach()
.cpu()
.numpy()
)
self.log(
"avg_loss_val_epoch",
e_v_avg_loss.item(),
on_step=False,
on_epoch=True,
sync_dist=True,
)
e_v_avg_loss_cl = (
torch.stack([x["loss_cl"] for x in self.validation_step_outputs])
.mean()
.detach()
.cpu()
.numpy()
)
self.log(
"avg_loss_cl_val_epoch",
e_v_avg_loss_cl.item(),
on_step=False,
on_epoch=True,
sync_dist=True,
)
e_v_avg_loss_mlm = (
torch.stack([x["loss_mlm"] for x in self.validation_step_outputs])
.mean()
.detach()
.cpu()
.numpy()
)
self.log(
"avg_loss_mlm_val_epoch",
e_v_avg_loss_mlm.item(),
on_step=False,
on_epoch=True,
sync_dist=True,
)
print(
"val_epoch:",
self.current_epoch,
"avg_loss:",
e_v_avg_loss,
"avg_cl_loss:",
e_v_avg_loss_cl,
"avg_mlm_loss:",
e_v_avg_loss_mlm,
)
self.validation_step_outputs.clear()
def configure_optimizers(self):
# Optimizer
self.trainable_params = [
param for param in self.parameters() if param.requires_grad
]
optimizer = AdamW(self.trainable_params, lr=self.lr)
# Scheduler
warmup_steps = self.n_batches // 3
total_steps = self.n_batches * self.n_epochs - warmup_steps
scheduler = get_linear_schedule_with_warmup(
optimizer, warmup_steps, total_steps
)
return [optimizer], [scheduler]
|