# args.embedding_dim,
# args.rnn_dim,
# args.answer_vocab_size, args.fixed_embed)
optimizer = optim.Adam(model.parameters(), lr=2.5 * 1e-4)
start_epoch = 0
with open(os.path.join(model_dir, 'model.pkl'), 'wb') as f:
pickle.dump(model, f)
with open(os.path.join(model_dir, 'optimizer.pkl'), 'wb') as f:
pickle.dump(optimizer, f)
if args.multi_gpu:
model = nn.DataParallel(model, device_ids=args.multi_gpu)
model.to(device)
with open(os.path.join(qa_dir, 'idx_word_dict.pkl'), 'rb') as f:
idx_to_word_dict = pickle.load(f)
idx_to_question = idx_to_word_dict['idx_to_question']
idx_to_question_type = idx_to_word_dict['idx_to_question_type']
idx_to_answer = idx_to_word_dict['idx_to_answer']
is_bert = args.model == 'base_bert'
train_loader, test_loader, input_dim = dataset.load_data(
args.datasetname, args.batch_size * multiplier, args.input_dim, is_bert,
multiplier)
writer = SummaryWriter(log_dir)
for epoch in range(1 + start_epoch, args.epochs + 1 + start_epoch):
