Can I use CPU mem to save GPU mem at sequential model?

nlp
2

You could check the CPU offloading for your use case.