# collate_fn=collate_fn_vi,
# shuffle=True)
val_loader_vi = torch.utils.data.DataLoader(dataset=val_vi,
batch_size=batch_size,
collate_fn=collate_fn_vi,
shuffle=True)
# val_loader_vi = torch.utils.data.DataLoader(dataset=val_vi,
# batch_size=batch_size,
# collate_fn=collate_fn_vi,
# shuffle=True)
test_loader_vi = torch.utils.data.DataLoader(dataset=test_vi,
batch_size=batch_size,
collate_fn=collate_fn_vi,
shuffle=True)
encoder = EncoderRNN(lang_vi.n_words, hidden_size, n_layers,
dropout=dropout).to(device)
decoder = DecoderRNN(hidden_size,
lang_en_vi.n_words,
n_layers,
dropout=dropout).to(device)
# Initialize optimizers and criterion
encoder_optimizer = optim.Adam(encoder.parameters(), lr=learning_rate)
decoder_optimizer = optim.Adam(decoder.parameters(),
lr=learning_rate * decoder_learning_ratio)
criterion = nn.CrossEntropyLoss().to(device)
# Keep track of time elapsed and running averages
start = time.time()