princeton-nlp / SimCSE

[EMNLP 2021] SimCSE: Simple Contrastive Learning of Sentence Embeddings https://arxiv.org/abs/2104.08821

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

关于 Supervised SimCSE 的 GPU Memory Usage

AnonymXXXXX opened this issue · comments

Hi~

根据论文附录,有监督场景下(使用 NLI 数据集)SimCSE-BERT-base 和 SimCSE-RoBERTa-base 的 batch size 为 512,seq len 为 32。请问这样的训练配置需要多少显存(GPU Memory Usage)?以及 SimCSE-BERT-base 和 SimCSE-RoBERTa-base 的显存占用是否一致?(需要据此确定自己的硬件环境,谢谢!)

Hi,

The unsupervised model was trained with 1 3090 and the supervised model was trained with 4x3090.

Ok, that should be 24G*4, thanks for your reply.