Opennmt-py: RuntimeError: Trying to backward through the graph a second time

[xhzhao]

i came across the following error with pytorch 0.4.0
(commit id : 0f198fa7239fff38f0d9e21025faa19b7d9ff466):

Loading train dataset from data/demo.train.1.pt, number of examples: 10000
/home/zhaoxiao/test/OpenNMT-py/onmt/modules/GlobalAttention.py:176: UserWarning: Implicit dimension choice for softmax has been deprecated. Change the call to include dim=X as an argument.
align_vectors = self.sm(align.view(batch*targetL, sourceL))
/home/zhaoxiao/anaconda3-cpu/envs/pytorch-040/lib/python3.6/site-packages/torch/nn/modules/container.py:91: UserWarning: Implicit dimension choice for log_softmax has been deprecated. Change the call to include dim=X as an argument.
input = module(input)
/home/zhaoxiao/test/OpenNMT-py/onmt/Loss.py:144: UserWarning: invalid index of a 0-dim tensor. This will be an error in PyTorch 0.5. Use tensor.item() to convert a 0-dim tensor to a Python number
return onmt.Statistics(loss[0], non_padding.sum(), num_correct)
Traceback (most recent call last):
File "train.py", line 438, in
main()
File "train.py", line 430, in main
train_model(model, fields, optim, data_type, model_opt)
File "train.py", line 252, in train_model
train_stats = trainer.train(train_iter, epoch, report_func)
File "/home/zhaoxiao/test/OpenNMT-py/onmt/Trainer.py", line 178, in train
report_stats, normalization)
File "/home/zhaoxiao/test/OpenNMT-py/onmt/Trainer.py", line 311, in _gradient_accumulation
trunc_size, self.shard_size, normalization)
File "/home/zhaoxiao/test/OpenNMT-py/onmt/Loss.py", line 124, in sharded_compute_loss
loss.div(normalization).backward()
File "/home/zhaoxiao/anaconda3-cpu/envs/pytorch-040/lib/python3.6/site-packages/torch/autograd/variable.py", line 120, in backward
torch.autograd.backward(self, gradient, retain_graph, create_graph)
File "/home/zhaoxiao/anaconda3-cpu/envs/pytorch-040/lib/python3.6/site-packages/torch/autograd/__init__.py", line 81, in backward
variables, grad_variables, retain_graph, create_graph)
RuntimeError: Trying to backward through the graph a second time, but the buffers have already been freed. Specify retain_graph=True when calling backward the first time.

[srush]

0.4 is not released yet, so currently we only support 0.3. We will try to move to 0.4 when it nears release.