thunlp
Improvement can be made on optimizing GPU memory occupation
Some variables can be deleted to optimize GPU memory use. For example, in the implementation of TransR, since the parameters pretrained on TransE are set to TransR, the tensors occupied by TransE can be released by deleting the TransE model and cleaning CUDA cache. According to my test, this can save significant GPU memory space when it comes to large datasets.
Some variables can be deleted to optimize GPU memory use. For example, in the implementation of TransR, since the parameters pretrained on TransE are set to TransR, the tensors occupied by TransE can be released by deleting the TransE model and cleaning CUDA cache. According to my test, this can save significant GPU memory space when it comes to large datasets.
Hey, I think we can communicate with openKE, the author doesn't reply for a long time, my qq is 1575797665
