Error while running encoder/play.ipynb

[LeenaShekhar]

When I run the following lines:

embeddings = model.encode(sentences, bsize=128, tokenize=False, verbose=True)
print('nb sentences encoded : {0}'.format(len(embeddings)))

I get the following error:

Nb words kept : 128201/130068 (98.56 %)

RuntimeError Traceback (most recent call last)
in ()
----> 1 embeddings = model.encode(sentences, bsize=128, tokenize=False, verbose=True)
2 print('nb sentences encoded : {0}'.format(len(embeddings)))

/home/leena/Downloads/InferSent-master/encoder/models.py in encode(self, sentences, bsize, tokenize, verbose)
177 if self.use_cuda:
178 batch = batch.cuda()
--> 179 batch = self.forward((batch, lengths[stidx:stidx + bsize])).data.cpu().numpy()
180 embeddings.append(batch)
181 embeddings = np.vstack(embeddings)

/home/leena/Downloads/InferSent-master/encoder/models.py in forward(self, sent_tuple)
48
49 # Un-sort by length
---> 50 idx_unsort = torch.from_numpy(idx_unsort).cuda() if self.use_cuda else torch.from_numpy(idx_sort)
51 sent_output = sent_output.index_select(1, Variable(idx_unsort))
52

RuntimeError: from_numpy expects an np.ndarray but got torch.LongTensor

Version details:
'3.6.0 |Anaconda custom (64-bit)| (default, Dec 23 2016, 12:22:00) \n[GCC 4.4.7 20120313 (Red Hat 4.4.7-1)]'
In [ ]:
torch-0.1.12.post2

[wasiahmad]

There is a small bug. Line no. 50 should be:

idx_unsort = torch.from_numpy(idx_unsort).cuda() if self.use_cuda else torch.from_numpy(idx_unsort)

instead of,

idx_unsort = torch.from_numpy(idx_unsort).cuda() if self.use_cuda else torch.from_numpy(idx_sort)

In the else condition, it should be torch.from_numpy(idx_unsort) rather than torch.from_numpy(idx_sort). idx_sort is a torch.LongTensor. See line no. 41.

[aconneau]

Indeed, thanks Wasi. Solved in 9357c2b (bug coming from recent commit 260bfd4).

[LeenaShekhar]

This is working now. Thank you.