xiaolinandy / rods Goto Github PK

你好，除了论文中提到的CSDS和MC数据集，是否还有其他中文对话/会议摘要数据集呢？尤其是会议摘要（长对话）

PGN_interact

您好，我最近基于CSDS的数据集进行实验，和您的实验结果进行了对比。但是我们发现基于BERT的模型的rouge得分要比您论文中的结果高很多。我们也对您提供的结果文件进行了评价，发现结果也是比您论文里的结果高很多。评价的方式和您代码中的一样。想问一下，您在论文里报道的结果是不是并不是最优的结果。
我们对您提供的结果文件评价的结果：
user summary:
58.88/43.26/56.80
agent summary:
53.40/38.65/51.10

Hello, when I run the BERTAbs interact experiment, run_CSDS.sh , I encountered a problem, as follows:

Traceback (most recent call last):
File "train.py", line 131, in
train_abs(args, device_id)
File "/mnt/ssd/lcq/RODS/models/BERT_interact/train_abstractive.py", line 274, in train_abs
train_abs_single(args, device_id)
File "/mnt/ssd/lcq/RODS/models/BERT_interact/train_abstractive.py", line 335, in train_abs_single
trainer.train(train_iter_fct, args.train_steps)
File "/mnt/ssd/lcq/RODS/models/BERT_interact/models/trainer.py", line 162, in train
report_stats)
File "/mnt/ssd/lcq/RODS/models/BERT_interact/models/trainer.py", line 249, in _gradient_accumulation
outputs, _, scores = self.model(src, tgts, segs, clss, mask_src, mask_tgts, mask_cls, self.args.merge, self.args.inter_weight, role_mask)
File "/home/ubuntu/anaconda3/envs/rods/lib/python3.7/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
result = self.forward(*input, **kwargs)
File "/mnt/ssd/lcq/RODS/models/BERT_interact/models/model_builder.py", line 274, in forward
user_outputs, agent_outputs, state_user, state_agent, user_scores, agent_scores, kl_mask_user, kl_mask_agent = self.role_decoder(tgts[0][:, :-1], tgts[1][:, :-1], top_vec, user_dec_state, agent_dec_state, merge_type, inter_weight, role_mask)
File "/home/ubuntu/anaconda3/envs/rods/lib/python3.7/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
result = self.forward(*input, **kwargs)
File "/mnt/ssd/lcq/RODS/models/BERT_interact/models/decoder.py", line 477, in forward
step=step)
File "/home/ubuntu/anaconda3/envs/rods/lib/python3.7/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
result = self.forward(*input, **kwargs)
File "/mnt/ssd/lcq/RODS/models/BERT_interact/models/decoder.py", line 186, in forward
type="context")
File "/home/ubuntu/anaconda3/envs/rods/lib/python3.7/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
result = self.forward(*input, **kwargs)
File "/mnt/ssd/lcq/RODS/models/BERT_interact/models/neural.py", line 341, in forward
batch_size = key.size(0)
AttributeError: 'str' object has no attribute 'size'

I try to print the 'key'(batch_size = key.size(0)) in the neural.py, the last output is not tensor, but the string "last_hidden_state"

如题