SPINN example not batching and never testing fold

[fac2003]

The SPINN example reads:

    for batch_idx, batch in enumerate(train_iter):
        opt.zero_grad()

        all_logits, all_labels = [], []
        fold = torchfold.Fold(device=device)
        # TODO: incorrect logic here, the for loop goes through the entire dataset, not the batch:
        for example in batch.dataset:

The line for example in batch.dataset: does not iterate though a batch. Instead, it iterates through an entire training set. This means that the fold code after the loop will not run until after the entire training set is scanned.
If I accumulate up to batch_size examples to test the fold code after the loop, I get an exception:

File "/Users/fac2003/PycharmProjects/torchfold/examples/snli/spinn-example.py", line 54, in leaf
embedded = self.embeddings(word_id)
File "/Users/fac2003/pytorch_env/lib/python3.6/site-packages/torch/nn/modules/module.py", line 491, in call
result = self.forward(*input, **kwargs)
File "/Users/fac2003/pytorch_env/lib/python3.6/site-packages/torch/nn/modules/sparse.py", line 108, in forward
self.norm_type, self.scale_grad_by_freq, self.sparse)
File "/Users/fac2003/pytorch_env/lib/python3.6/site-packages/torch/nn/functional.py", line 1076, in embedding
return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)
RuntimeError: index out of range at /Users/soumith/code/builder/wheel/pytorch-src/aten/src/TH/generic/THTensorMath.c:343

Presumably, the word id is larger than the max number of words. This is likely a consequence of initializing the tree with a non vectorized example. I am wondering why the torchtext tensors in batch are not used instead (they have the correct batch dimensionality).