训练时内存持续增长到OOM

Determined22 / zh-NER-TF

A very simple BiLSTM-CRF model for Chinese Named Entity Recognition 中文命名实体识别 (TensorFlow)

2.32k stars 937 forks source link

Open xiaofei05 opened 5 years ago

xiaofei05 commented 5 years ago

用的自己的数据，数据量比较大，训练的时候大概batch_size=128 2000 step时出现OOM，请问是哪里出现问题了呢

GanjinZero commented 5 years ago

同出现此问题

GanjinZero commented 5 years ago

把main.py中的这一行改了一下

config.gpu_options.per_process_gpu_memory_fraction = 0.9

就好了