Vipning issues

Results 8 issues of


                                            Vipning

OOM ReourceExhaustedError

Caused by op 'decoder/num_blocks_5/vanilla_attention/scaled_dot_product_attention/dropout/dropout/random_uniform/RandomUniform' ResourceExhaustedError (see above for traceback): OOM when allocating tensor with shape[1024,98,98] and type float on /job:localhost/replica:0/task:0/device:CPU:0 by allocator cpu [[Node: decoder/num_blocks_5/vanilla_attention/scaled_dot_product_attention/dropout/dropout/random_uniform/RandomUniform = RandomUniform[T=DT_INT32, dtype=DT_FLOAT, seed=0, seed2=0,...

K，Q，V映射

我想问一下多头attention那里，通过一次线性映射产生K,Q,V，paper中说要8次不同的映射。代码实现时，是因为，8次不同的映射可以一次完成，只要权重参数的数量已经相当于8次映射的数量了么？

I want to konw the way to preprocess CoNLL 2012 dataset,could you give some instructor?

想问一下ontonote5.0中propBank是不是能当成treebank用？

how to understand the head_word_emb in the initial file ,glove_50_300_2.txt,how to get the file?

batch_size 可以不设为1么？

请问，运行的环境配置是什么？

数据采集

运行了你的项目非常优秀，你的爬取的数据是存在monogoDB里的么？怎么变成了json数据？