Release the memory / GPU memory after usage.

[GalaxieT]

Hi!

I was trying to get pretrained embeddings by encoding some sentence, but I noticed that the memory and GPU memory would not be released after encoding and would be kept until the whole program ended (even if the next parts are not relavant to the pretrained model). So I wonder is there any method to inform the model to release all the memory resources?

[bandoos]

Hello :) Anyone has news/suggestions on this problem?

[aadityabhatia]

Have you tried using start_multi_process_pool and then stop_multi_process_pool?

Ref: https://www.sbert.net/docs/package_reference/SentenceTransformer.html