def do_train(args):
device = paddle.set_device(args.select_device)
# Define dataloader
train_loader, eval_loader, src_vocab_size, tgt_vocab_size, eos_id = create_train_loader(
args)
model = paddle.Model(
Seq2SeqAttnModel(src_vocab_size, tgt_vocab_size, args.hidden_size,
args.hidden_size, args.num_layers, args.dropout,
eos_id))
grad_clip = nn.ClipGradByGlobalNorm(args.max_grad_norm)
optimizer = paddle.optimizer.Adam(learning_rate=args.learning_rate,
parameters=model.parameters(),
grad_clip=grad_clip)
ppl_metric = Perplexity()
model.prepare(optimizer, CrossEntropyCriterion(), ppl_metric)
print(args)
if args.init_from_ckpt:
model.load(args.init_from_ckpt)
print("Loaded checkpoint from %s" % args.init_from_ckpt)
model.fit(train_data=train_loader,
eval_data=eval_loader,
epochs=args.max_epoch,
eval_freq=1,
save_freq=1,
save_dir=args.model_path,
log_freq=args.log_freq)
def do_train(args):
device = paddle.set_device("gpu" if args.use_gpu else "cpu")
# Define dataloader
train_loader, vocab_size, pad_id = create_train_loader(args.batch_size)
model = paddle.Model(
Seq2SeqAttnModel(vocab_size, args.hidden_size, args.hidden_size,
args.num_layers, pad_id))
optimizer = paddle.optimizer.Adam(
learning_rate=args.learning_rate, parameters=model.parameters())
ppl_metric = Perplexity()
model.prepare(optimizer, CrossEntropyCriterion(), ppl_metric)
print(args)
model.fit(train_data=train_loader,
epochs=args.max_epoch,
eval_freq=1,
save_freq=1,
save_dir=args.model_path,
log_freq=args.log_freq,
callbacks=[paddle.callbacks.VisualDL('./log')])
def do_train(args):
device = paddle.set_device(args.device)
# Define dataloader
train_loader, vocab = create_train_loader(args.batch_size)
vocab_size = len(vocab)
pad_id = vocab[vocab.eos_token]
model = paddle.Model(
Seq2SeqAttnModel(vocab_size, args.hidden_size, args.hidden_size,
args.num_layers, pad_id))
optimizer = paddle.optimizer.Adam(learning_rate=args.learning_rate,
parameters=model.parameters())
ppl_metric = Perplexity()
model.prepare(optimizer, CrossEntropyCriterion(), ppl_metric)
print(args)
model.fit(train_data=train_loader,
epochs=args.max_epoch,
eval_freq=1,
save_freq=1,
save_dir=args.model_path,
log_freq=args.log_freq)