关于GPU的数量。 - Githubissues

nanguoshun / LSR

Pytorch Implementation of our ACL 2020 Paper "Reasoning with Latent Structure Refinement for Document-Level Relation Extraction"

127 stars 22 forks source link

Closed yangxia605 closed 4 years ago

yangxia605 commented 4 years ago

您好：首先，感谢您出色的工作并且分享代码。请问一下您跑这个实验用了几块GPU呢？感谢！

nanguoshun commented 4 years ago

@yangxia605 谢谢关注！Glove+LSR，BZ=20需要2个RTX2080 (2*11GB)，如果batchsize设置小的话，也可以在一个卡 (11G) 上跑

qiunlp commented 4 years ago

您好：没有GPU，如果不计较速度，是否也可以在CPU上跑？感谢！

nanguoshun commented 4 years ago

@qiunlp 不太建议用CPU跑，会非常慢，因为用GPU的话（Glove+LSR），Training大概需要8个小时

nanguoshun commented 4 years ago

I close this issue in case that there are no further new queries.