kermitt2 / delft

a Deep Learning Framework for Text

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Bad performance to produce ELMo embedding

SeanLee97 opened this issue · comments

commented

In my setting: batch size = 128, max_sequence_length = 300. It really takes time to produce the ELMo embedding, about 3s per batch. Is this a normal time?

commented

I have solved this problem. This is because my mounted driver has a poor writing speed rate. When I set cache-training: false, the problem was solved.

Using the ELMo embedding cache makes sense only with good SSD indeed.