def start():
# 优先使用缓存
if not os.path.exists(args.TRAIN) or not os.path.exists(args.VALID):
produce_data(user_define=USER_DEFINE)
if os.path.exists(args.TRAIN_CACHE):
train_iter, num_train_steps = torch.load(args.TRAIN_CACHE)
else:
train_iter, num_train_steps = create_batch_iter("train")
if os.path.exists(args.VALID_CACHE):
eval_iter = torch.load(args.VALID_CACHE)
else:
eval_iter = create_batch_iter("dev")
epoch_size = num_train_steps * args.train_batch_size * args.gradient_accumulation_steps / args.num_train_epochs
pbar = ProgressBar(epoch_size=epoch_size, batch_size=args.train_batch_size)
model = Bert_CRF.from_pretrained(args.bert_model, num_tag=len(args.labels))
for name, param in model.named_parameters():
if param.requires_grad:
print(name)
fit(model=model,
training_iter=train_iter,
eval_iter=eval_iter,
num_epoch=args.num_train_epochs,
pbar=pbar,
num_train_steps=num_train_steps,
verbose=1)
def start():
parser = argparse.ArgumentParser()
parser.add_argument(
"--do_not_train_ernie",
default=False,
action='store_true',
)
parser.add_argument(
"--do_CRF",
default=False,
action='store_true',
)
arg = parser.parse_args()
args.do_not_train_ernie = arg.do_not_train_ernie
args.do_CRF = arg.do_CRF
produce_data()
train_iter, num_train_steps = create_batch_iter("train")
eval_iter = create_batch_iter("dev")
epoch_size = num_train_steps * args.train_batch_size * args.gradient_accumulation_steps / args.num_train_epochs
pbar = ProgressBar(epoch_size=epoch_size, batch_size=args.train_batch_size)
if args.load_weight:
model = load_model(args.output_dir)
else:
model = Bert_CRF.from_pretrained(args.bert_model,
num_tag=len(args.labels))
for name, param in model.named_parameters():
if param.requires_grad:
print(name)
fit(model=model,
training_iter=train_iter,
eval_iter=eval_iter,
num_epoch=args.num_train_epochs,
pbar=pbar,
num_train_steps=num_train_steps,
verbose=1)
def start():
produce_data()
train_iter, num_train_steps = create_batch_iter("train")
eval_iter = create_batch_iter("dev")
epoch_size = num_train_steps * args.train_batch_size * args.gradient_accumulation_steps / args.num_train_epochs
pbar = ProgressBar(epoch_size=epoch_size, batch_size=args.train_batch_size)
model = Bert_CRF.from_pretrained(args.bert_model, num_tag=len(args.labels))
for name, param in model.named_parameters():
if param.requires_grad:
print(name)
fit(model=model,
training_iter=train_iter,
eval_iter=eval_iter,
num_epoch=args.num_train_epochs,
pbar=pbar,
num_train_steps=num_train_steps,
verbose=1)
from Io.data_loader import create_batch_iter
from preprocessing.data_processor import produce_data
import torch
import os
import json
import config.args as args
from util.model_util import load_model
args.do_inference = True
produce_data()
test_iter = create_batch_iter("inference")
epoch_size = args.train_batch_size * args.gradient_accumulation_steps / args.num_train_epochs
model = load_model(args.output_dir)
num_epoch = args.num_train_epochs
device = torch.device(
args.device if torch.cuda.is_available() and not args.no_cuda else "cpu")
param_optimizer = list(model.named_parameters())
no_decay = ['bias', 'LayerNorm.bias', 'LayerNorm.weight']
optimizer_grouped_parameters = [{
'params':
[p for n, p in param_optimizer if not any(nd in n for nd in no_decay)],
'weight_decay':
0.01
}, {
'params':
[p for n, p in param_optimizer if any(nd in n for nd in no_decay)],
'weight_decay':