eval_dataset,
batch_size=args.test_batch_size,
sampler=eval_sampler,
num_workers=args.data_workers,
collate_fn=batchify_features_for_test,
pin_memory=args.cuda,
)
# -------------------------------------------------------------------------------------------
# Preprare Model & Optimizer
# -------------------------------------------------------------------------------------------
logger.info(
" ************************** Initilize Model ************************** "
)
try:
model, checkpoint_epoch = KeyphraseSpanExtraction.load_checkpoint(
args.eval_checkpoint, args)
model.set_device()
except ValueError:
print("Could't Load Pretrain Model %s" % args.eval_checkpoint)
if args.local_rank == 0:
torch.distributed.barrier()
if args.n_gpu > 1:
model.parallelize()
if args.local_rank != -1:
model.distribute()
# -------------------------------------------------------------------------------------------
# Method Select
train_data_loader
) // args.gradient_accumulation_steps * args.max_train_epochs
# -------------------------------------------------------------------------------------------
# Preprare Model & Optimizer
# -------------------------------------------------------------------------------------------
# 7.初始化模型和优化器
logger.info(
" ************************** Initialize Model & Optimizer ************************** "
)
"""
`args.checkpoint_file`, loaded checkpoint model continue training.
`args.load_checkpoint`, default=False
"""
if args.load_checkpoint and os.path.isfile(args.checkpoint_file):
model = KeyphraseSpanExtraction.load_checkpoint(
args.checkpoint_file, args)
else:
logger.info('Training model from scratch...')
model = KeyphraseSpanExtraction(args)
model.init_optimizer(num_total_steps=t_total)
if args.local_rank == 0:
torch.distributed.barrier()
model.set_device()
if args.n_gpu > 1:
model.parallelize()
if args.local_rank != -1:
model.distribute()
t_total = (len(train_data_loader) // args.gradient_accumulation_steps *
args.max_train_epochs)
# -------------------------------------------------------------------------------------------
# Preprare Model & Optimizer
# -------------------------------------------------------------------------------------------
if args.local_rank not in [-1, 0]:
torch.distributed.barrier(
) # Make sure only the first process in distributed training will download model & vocab
logger.info(
" ************************** Initilize Model & Optimizer ************************** "
)
if args.load_checkpoint and os.path.isfile(args.checkpoint_file):
model, checkpoint_epoch = KeyphraseSpanExtraction.load_checkpoint(
args.checkpoint_file, args)
else:
logger.info("Training model from scratch...")
model = KeyphraseSpanExtraction(args)
# initial optimizer
model.init_optimizer(num_total_steps=t_total)
# -------------------------------------------------------------------------------------------
if args.local_rank == 0:
torch.distributed.barrier(
) # Make sure only the first process in distributed training will download model & vocab
# -------------------------------------------------------------------------------------------
# set model device
model.set_device()