def load_model(config, num_train_steps, label_list):
device = torch.device('cuda')
n_gpu = config.n_gpu
model = BertQueryNER(config, )
if config.n_gpu > 1:
model = torch.nn.DataParallel(model)
model.to(device)
# prepare optimzier
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":
0.0
}]
optimizer = AdamW(optimizer_grouped_parameters,
lr=config.learning_rate,
eps=10e-8)
sheduler = None
return model, optimizer, sheduler, device, n_gpu
def load_model(config, num_train_steps, label_list, logger):
device = torch.device("cuda")
n_gpu = config.n_gpu
model = BertQueryNER(config, train_steps=num_train_steps)
model.to(device)
if config.n_gpu > 1:
model = torch.nn.DataParallel(model)
# prepare optimzier
param_optimizer = list(model.named_parameters())
no_decay = ["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": config.weight_decay},
{"params": [p for n, p in param_optimizer if any(nd in n for nd in no_decay)], "weight_decay": 0.0}]
optimizer = build_fp32_optimizer(config, optimizer_grouped_parameters,)
scheduler = build_lr_scheduler(config, optimizer, num_train_steps)
if config.fp16:
try:
from apex import amp
except ImportError:
raise ImportError("Please install apex from https://www.github.com/nvidia/apex to use fp16 training.")
model, optimizer = amp.initialize(model, optimizer, opt_level=config.amp_level)
# Distributed training (should be after apex fp16 initialization)
if config.local_rank != -1 and config.data_parallel == "ddp":
model = torch.nn.parallel.DistributedDataParallel(
model, device_ids=[config.local_rank], output_device=config.local_rank, find_unused_parameters=True
)
return model, optimizer, scheduler, device, n_gpu
def load_model(config, num_train_steps, label_list):
device = torch.device("cuda")
n_gpu = config.n_gpu
model = BertQueryNER(config, )
model.to(device)
if config.n_gpu > 1:
model = torch.nn.DataParallel(model)
# prepare optimzier
param_optimizer = list(model.named_parameters())
no_decay = ["bias", "LayerNorm.bias", "LayerNorm.weight", 'gamma', 'beta']
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": 0.0}]
optimizer = AdamW(optimizer_grouped_parameters, lr=config.learning_rate, betas=(0.9, 0.98), eps=1e-6, weight_decay=0.01)
# optimizer = BertAdam(optimizer_grouped_parameters, lr=config.learning_rate, warmup=config.warmup_proportion,
# t_total=num_train_steps, max_grad_norm=config.clip_grad)
sheduler = None
if config.fp16:
try:
from apex import amp
except ImportError:
raise ImportError("Please install apex from https://www.github.com/nvidia/apex to use fp16 training.")
model, optimizer = amp.initialize(model, optimizer, opt_level=config.amp_level)
# Distributed training (should be after apex fp16 initialization)
if config.local_rank != -1:
model = torch.nn.parallel.DistributedDataParallel(
model, device_ids=[config.local_rank], output_device=config.local_rank, find_unused_parameters=True
)
return model, optimizer, sheduler, device, n_gpu
def load_model(config, label_list):
device = torch.device("cuda")
n_gpu = config.n_gpu
model = BertQueryNER(config)
checkpoint = torch.load(config.saved_model)
model.load_state_dict(checkpoint)
model.to(device)
if config.n_gpu > 1:
model = torch.nn.DataParallel(model)
return model, device, n_gpu