search

travel-go / Abstractive-Text-Summarization

Contrastive Attention Mechanism for Abstractive Text Summarization
Other
40 stars 7 forks source link

version of fairseq #4

Open cedar33 opened 4 years ago

cedar33 commented 4 years ago

when I run train.py, an error occured and said 'no module named data', then I copied the data module from pytorch's fairseq(version 0.8), more errors come out, which fairseq`s version in your code?

travel-go commented 4 years ago

The fairseq's version is 0.5.0 and I have uploaded the data module.

cedar33 commented 4 years ago

The fairseq's version is 0.5.0 and I have uploaded the data module.

thanks, I have get this after 8h training, however the summary is unsatisfactory.the dataset is LCSTS dataset. | epoch 006: 100%|▉| 11890/11899 [1:20:56<00:03, 2.39it/s, loss=6.979, ppl=126.17, wps=9312, ups=2.4, wpb=3803, bsz=202, num_updates=71385, lr=0.00025