brightmart / albert_zh

A LITE BERT FOR SELF-SUPERVISED LEARNING OF LANGUAGE REPRESENTATIONS, 海量中文预训练ALBERT模型
https://arxiv.org/pdf/1909.11942.pdf
3.94k stars 753 forks source link

tesla V100 跑xlarge模型OOM #119

Open hejunqing opened 4 years ago

hejunqing commented 4 years ago

介绍中说xlarge模型只有bert_base模型 二分之一大小,bert_base只需要10G的显卡,而xlarge模型在V100上竟然OOM,请问是不是模型有问题?xlarge模型需要多大的显存?

JQIANG125 commented 4 years ago

oom和batch size和length有关,我p100可以跑32batch 64length,再多就可能出问题