warnings.warn('model_parameters is not set. '
'Randomly initialized model will be used for inference.')
get_pretrained = True
get_model_params = {
'name': args.model_name,
'dataset_name': args.dataset,
'pretrained': get_pretrained,
'ctx': ctxs,
'use_decoder': False,
'dropout': args.dropout,
'attention_dropout': args.attention_dropout
}
xlnet_base, vocab, tokenizer = model.get_model(**get_model_params)
# initialize the rest of the parameters
initializer = mx.init.Normal(0.02)
do_regression = not task.class_labels
if do_regression:
num_classes = 1
loss_function = gluon.loss.L2Loss()
else:
num_classes = len(task.class_labels)
loss_function = gluon.loss.SoftmaxCELoss()
# reuse the XLnetClassifier class with num_classes=1 for regression
model = XLNetClassifier(xlnet_base,
units=xlnet_base._net._units,
dropout=0.1,
num_classes=num_classes)
type=int,
default=64,
help='Batch size for evaluation.')
parser.add_argument('--gpu', type=int, help='GPU id')
args = parser.parse_args()
start_time = time.time()
# Model
from transformer.model import get_model
with open(args.vocab_file, 'r') as f:
vocab = nlp.Vocab.from_json(f.read())
ctx = mx.gpu(args.gpu) if args.gpu is not None else mx.cpu()
model, vocab = get_model('transformerxl',
vocab=vocab,
dataset_name=args.dataset,
clamp_len=args.clamp_len)
model.initialize(ctx=ctx)
model.load_parameters(args.parameter_file, ignore_extra=False)
model.hybridize()
print(model)
# Data
if args.dataset == 'wt103':
val_dataset, test_dataset = [
nlp.data.WikiText103(segment=segment,
skip_empty=False,
bos=vocab.bos_token,
eos=vocab.eos_token)
for segment in ['val', 'test']
]