I tried torch.cuda.empty_cache() but no luck with diffusion pipeline. How can I clear video memory after an image generation?
From what I know…
-
You can kill a process:
1.1 !nvidia-smi (find your process [id])
1.2 !kill [id] -
By numba:
2.1 !pip install numba
2.2 from numba import cuda
2.3 device = cuda.get_current_device()
2.4 device.reset()