Problem about train

[Tickdack]

Summary

hello.

We want to use faiss to build an ivfpq index, the data size is about 200 million. When we train the index, we find that faiss needs to train the representative data once, which is a challenge for the memory and CPU of the machine, and the training time is also very long. Is there any way to batch the training/reduce the training time?

Thank you.

Platform

Centos 7

Running on:

• [ x ] CPU
• GPU

Interface:

• [ x ] C++
• Python