The checkpoint model does not generate alignment

[xiulinyang]

Hi, I tried to use the unguided checkpoint model and the command you provided to generate German AMR alignments but it just copied the input in the prediction folder I set. Below is the log message.

Also, there might be a typo in the README: the version of transformers should be > 3.0.

Many thanks in advance!

/local/xiulyang/anaconda3/envs/amralignment/lib/python3.8/site-packages/transformers/tokenization_utils_base.py:2211: FutureWarning: The `pad_to_max_length` argument is deprecated and will be removed in a future version, use `padding=True` or `padding='longest'` to pad to the longest sequence in the batch, or use `padding='max_length'` to pad to a max length. In this case, you can give a specific length with `max_length` (e.g. `max_length=45`) or leave max_length to None to pad to the maximal input size of the model (e.g. 512 for Bert). warnings.warn( WARNING:root:Invalid sentence! 0%| | 0/100 [00:00<?, ?it/s]{'input_ids': tensor([[ 0, 407, 449, 20905, 3999, 939, 5039, 364, 4272, 748, 36767, 12128, 2156, 885, 324, 952, 4394, 1943, 5079, 8797, 1437, 1725, 997, 2156, 1076, 29, 1437, 1725, 28, 118, 4969, 19596, 1180, 506, 21251, 5689, 364, 5101, 2084, 16793, 2156, 30864, 8554, 34679, 1794, 821, 2753, 242, 2420, 90, 885, 6831, 242, 479, 2], [ 0, 4594, 9876, 605, 2723, 9306, 13235, 4832, 22, 9938, 20399, 90, 295, 1725, 1872, 479, 22, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], [ 0, 10915, 1725, 858, 13235, 364, 179, 1811, 2001, 479, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], [ 0, 8318, 2156, 19958, 1437, 1725, 2084, 1610, 2156, 16, 90, 70, 293, 842, 5039, 449, 22593, 479, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], [ 0, 9938, 16, 90, 14001, 30864, 1023, 27785, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], [ 0, 12611, 18965, 25666, 14839, 225, 1690, 12614, 620, 4279, 17487, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], [ 0, 20963, 449, 8615, 620, 4279, 69, 2156, 449, 459, 5101, 9508, 462, 17487, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]], device='cuda:0'), 'attention_mask': tensor([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]], device='cuda:0')} /pytorch/aten/src/ATen/native/BinaryOps.cpp:81: UserWarning: Integer division of tensors using div or / is deprecated, and in a future release div will perform true division as in Python 3. Use true_divide or floor_divide (// in Python) instead. ['Ġ</s>', 'Ġ<s>', 'W', 'ir', 'Ġha', 'ben', 'Ġe', 'uch', 'Ġe', 'uch', 'Ġe', 'uch', 'Ġv', 'orst', 'ellen', 'Ġ,', 'Ġw', 'ie', 'ĠÃ', 'Ġ</s>'] 0%| | 0/100 [00:00<?, ?it/s]

[Carlosml26]

Hi there,

The issue is that the checkpoint released is the unguided model for AMR parsing in English (it is based in Bart). To move to other languages, you need a checkpoint trained in a cross-lingual fashion using mBart. We will release it soon, sorry for any inconvenience.