def build_helper(gz, args): return DataHelper(args.training_samples, gz=gz, config=args)
for i, l in enumerate(total_train_loss):
print("Avg-LOSS{}/batch/step: {}".format(i, l / VERBOSE_STEP))
total_train_loss = [0] * 5
if __name__ == "__main__":
parser = argparse.ArgumentParser()
args = set_config()
args.n_gpu = torch.cuda.device_count()
if args.seed == 0:
args.seed = random.randint(0, 100)
set_seed(args)
helper = DataHelper(gz=True, config=args)
args.n_type = helper.n_type # 2
# Set datasets
Full_Loader = helper.train_loader
# Subset_Loader = helper.train_sub_loader
dev_example_dict = helper.dev_example_dict
dev_feature_dict = helper.dev_feature_dict
eval_dataset = helper.dev_loader
roberta_config = BC.from_pretrained(args.bert_model)
encoder = BertModel.from_pretrained(args.bert_model)
args.input_dim = roberta_config.hidden_size
model = BertSupportNet(config=args, encoder=encoder)
if args.trained_weight is not None:
model.load_state_dict(torch.load(args.trained_weight))