Hi, this is a follow-up to https://github.com/dmlc/xgboost/issues/3083 — given a suggestion to start a topic here, I’m doing so. I still experience this problem after compiling from trunk yesterday. My workaround is: model <- unserialize(serialize(model, connection=NULL)); gc()
, but maybe there’s a better way?
GPU memory not freed in R after training
Ah, thank you! It would probably make sense to link the new issue from the old one, as 3083 seems to be more visible in Google. I can’t do so as the issue is locked.