Holding GPU memory even after closing the browser

We are trying to provide a deep learning demo on Streamlit using GPUs. During each demo, we fine-tune deep neural models for user request. Therefore, each user gets different models, and they take up GPU memory. After we close a browser, the models still remain on the memory. Thus, whenever we start and run a new demo session, each one takes its memory and finally our server runs out-of-memory and the overall sessions are shut down. We would like to end a session and free the GPU memory whenever we close a browser running a session.
Is there any way?