hslcy / glossbert Goto Github PK

F.softmax is being called on the SequenceClassifierOutput object, not the tensor

Hi,

Thanks for the code. How are you dealing with class imbalance problem. There will be vastly more gloss sentence pair that have label 0 then that have label 1.

Can you please write steps on how to use the last checkpoint?

Hi! I was wondering in what order are yes/no appear in the model output. Is the first logit for yes or for no?

Hello,

are there plans to release pretrained models? The model is very expensive to train, and the research community would really benefit from having them available.

Thanks in advance.

您好，我看了Wordnet的网站，并没有找到这两个文件，请问这两个文件是从哪里得到的？可以说一下每一列的意义是什么吗？

感谢分享，想问一下文中的弱监督信号（引号等）有没有相关的理论依据和相似的研究？谢谢

Hello,
Would you tell me how much memory is needed for training GlossBERT using your example setting?
while I train the model, it always shows RuntimeError: CUDA out of memory.
My gpu is 1080Ti 10G.

error:
Traceback (most recent call last):
File "run_classifier_WSD_sent.py", line 706, in
main()
File "run_classifier_WSD_sent.py", line 520, in main
logits = model(input_ids=input_ids, token_type_ids=segment_ids, attention_mask=input_mask, labels=None)
File "/home/nlplab/tjwu/anaconda3/envs/GlossBERT/lib/python3.7/site-packages/torch/nn/modules/module.py", line 489, in call
result = self.forward(*input, **kwargs)
File "/home/nlplab/tjwu/anaconda3/envs/GlossBERT/lib/python3.7/site-packages/torch/nn/parallel/data_parallel.py", line 143, in forward
outputs = self.parallel_apply(replicas, inputs, kwargs)
File "/home/nlplab/tjwu/anaconda3/envs/GlossBERT/lib/python3.7/site-packages/torch/nn/parallel/data_parallel.py", line 153, in parallel_apply
return parallel_apply(replicas, inputs, kwargs, self.device_ids[:len(replicas)])
File "/home/nlplab/tjwu/anaconda3/envs/GlossBERT/lib/python3.7/site-packages/torch/nn/parallel/parallel_apply.py", line 83, in parallel_apply
raise output
File "/home/nlplab/tjwu/anaconda3/envs/GlossBERT/lib/python3.7/site-packages/torch/nn/parallel/parallel_apply.py", line 59, in _worker
output = module(*input, **kwargs)
File "/home/nlplab/tjwu/anaconda3/envs/GlossBERT/lib/python3.7/site-packages/torch/nn/modules/module.py", line 489, in call
result = self.forward(*input, **kwargs)
File "/home/nlplab/tjwu/GlossBERT/modeling.py", line 972, in forward
_, pooled_output = self.bert(input_ids, token_type_ids, attention_mask, output_all_encoded_layers=False)
File "/home/nlplab/tjwu/anaconda3/envs/GlossBERT/lib/python3.7/site-packages/torch/nn/modules/module.py", line 489, in call
result = self.forward(*input, **kwargs)
File "/home/nlplab/tjwu/GlossBERT/modeling.py", line 716, in forward
output_all_encoded_layers=output_all_encoded_layers)
File "/home/nlplab/tjwu/anaconda3/envs/GlossBERT/lib/python3.7/site-packages/torch/nn/modules/module.py", line 489, in call
result = self.forward(*input, **kwargs)
File "/home/nlplab/tjwu/GlossBERT/modeling.py", line 395, in forward
hidden_states = layer_module(hidden_states, attention_mask)
File "/home/nlplab/tjwu/anaconda3/envs/GlossBERT/lib/python3.7/site-packages/torch/nn/modules/module.py", line 489, in call
result = self.forward(*input, **kwargs)
File "/home/nlplab/tjwu/GlossBERT/modeling.py", line 380, in forward
attention_output = self.attention(hidden_states, attention_mask)
File "/home/nlplab/tjwu/anaconda3/envs/GlossBERT/lib/python3.7/site-packages/torch/nn/modules/module.py", line 489, in call
result = self.forward(*input, **kwargs)
File "/home/nlplab/tjwu/GlossBERT/modeling.py", line 338, in forward
self_output = self.self(input_tensor, attention_mask)
File "/home/nlplab/tjwu/anaconda3/envs/GlossBERT/lib/python3.7/site-packages/torch/nn/modules/module.py", line 489, in call
result = self.forward(*input, **kwargs)
File "/home/nlplab/tjwu/GlossBERT/modeling.py", line 298, in forward
attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))
RuntimeError: CUDA out of memory. Tried to allocate 384.00 MiB (GPU 0; 10.92 GiB total capacity; 10.20 GiB already allocated; 23.38 MiB free; 30.57 MiB cached)

We have checked the POS performance listed in Table 3 but the statistics do not add up. The overall performance is 77.6 (not 77.0) if we add all correct instances from each POS, shown in the following table, please check, THX.

Hi, have you run into the same problem?
I'm training on 8 16GB V100 GPU but got this exception (even only use 1):