The local backend keeps a Python process alive with the last models loaded. Call this to free the memory (GPU included). It restarts on the next generation call.
The local backend keeps a Python process alive with the last models loaded. Call this to free the memory (GPU included). It restarts on the next generation call.