help='batch size')
parser.add_argument('--target-seq-len',
type=int,
default=35,
help='sequence length')
parser.add_argument('--seed', type=int, default=1111, help='random seed')
parser.add_argument('--cuda', action='store_true', help='use CUDA')
parser.add_argument('--load',
type=str,
required=True,
help='where to load a model from')
args = parser.parse_args()
print(args)
init_seeds(args.seed, args.cuda)
print("loading model...")
lm = torch.load(args.load, map_location='cpu')
if args.cuda:
lm.cuda()
print(lm)
print("preparing data...")
tokenize_regime = 'words'
if args.characters:
tokenize_regime = 'chars'
ids = tokens_from_fn(args.data,
lm.vocab,
randomize=False,
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--train',
type=str,
required=True,
help='location of the train corpus')
parser.add_argument('--valid',
type=str,
required=True,
help='location of the valid corpus')
parser.add_argument(
'--characters',
action='store_true',
help='work on character level, whitespace is significant')
parser.add_argument('--shuffle-lines',
action='store_true',
help='shuffle lines before every epoch')
parser.add_argument('--batch-size',
type=int,
default=20,
metavar='N',
help='batch size')
parser.add_argument('--target-seq-len',
type=int,
default=35,
help='sequence length')
parser.add_argument('--lr',
type=float,
default=20,
help='initial learning rate')
parser.add_argument('--beta',
type=float,
default=0,
help='L2 regularization penalty')
parser.add_argument('--clip',
type=float,
default=0.25,
help='gradient clipping')
parser.add_argument('--epochs',
type=int,
default=40,
help='upper epoch limit')
parser.add_argument('--seed', type=int, default=1111, help='random seed')
parser.add_argument('--cuda', action='store_true', help='use CUDA')
parser.add_argument('--log-interval',
type=int,
default=200,
metavar='N',
help='report interval')
parser.add_argument('--val-interval',
type=int,
default=1000000,
metavar='N',
help='validation interval in number of tokens')
parser.add_argument('--val-report', help='where to put validation report')
parser.add_argument('--load',
type=str,
required=True,
help='where to load a model from')
parser.add_argument('--save',
type=str,
required=True,
help='path to save the final model')
args = parser.parse_args()
print(args)
init_seeds(args.seed, args.cuda)
print("loading model...")
lm = torch.load(args.load)
if args.cuda:
lm.cuda()
print(lm.model)
print("preparing data...")
tokenize_regime = 'words'
if args.characters:
tokenize_regime = 'chars'
train_ids = tokens_from_fn(args.train,
lm.vocab,
randomize=False,
regime=tokenize_regime)
train_batched = batchify(train_ids, args.batch_size, args.cuda)
train_data_tb = TemporalSplits(train_batched,
nb_inputs_necessary=lm.model.in_len,
nb_targets_parallel=args.target_seq_len)
train_data = TransposeWrapper(train_data_tb)
valid_ids = tokens_from_fn(args.valid,
lm.vocab,
randomize=False,
regime=tokenize_regime)
valid_batched = batchify(valid_ids, 10, args.cuda)
valid_data_tb = TemporalSplits(valid_batched,
nb_inputs_necessary=lm.model.in_len,
nb_targets_parallel=args.target_seq_len)
valid_data = TransposeWrapper(valid_data_tb)
def val_loss_fn(lm):
return evaluate_(lm, valid_data, use_ivecs=False, custom_batches=False)
initial_val_loss = val_loss_fn(lm)
print('Initial perplexity {:.2f}'.format(math.exp(initial_val_loss)))
print("training...")
lr = args.lr
best_val_loss = None
if args.val_report is not None:
val_report_f = open(args.val_report, 'w', buffering=1)
else:
val_report_f = sys.stdout
val_watcher = ValidationWatcher(lambda: val_loss_fn(lm), initial_val_loss,
args.val_interval, val_report_f)
optim = torch.optim.SGD(lm.parameters(), lr, weight_decay=args.beta)
for epoch in range(1, args.epochs + 1):
logger = ProgressLogger(epoch, args.log_interval, lr,
len(train_batched) // args.target_seq_len)
hidden = None
for X, targets in train_data:
X = X.t()
targets = targets.t().contiguous()
if hidden is None:
hidden = lm.model.init_hidden(args.batch_size)
hidden = repackage_hidden(hidden)
lm.train()
output, hidden = lm.model(X, hidden)
loss, nb_words = lm.decoder.neg_log_prob(output, targets)
loss /= nb_words
val_watcher.log_training_update(loss.data, nb_words)
optim.zero_grad()
loss.backward()
torch.nn.utils.clip_grad_norm(lm.parameters(), args.clip)
optim.step()
logger.log(loss.data)
val_loss = val_loss_fn(lm)
print(
epoch_summary(epoch, logger.nb_updates(),
logger.time_since_creation(), val_loss))
# Save the model if the validation loss is the best we've seen so far.
if not best_val_loss or val_loss < best_val_loss:
torch.save(lm, args.save)
best_val_loss = val_loss
else:
lr /= 2.0
pass