could I perhaps double check if below is a robust solution? it is working - However I does seem useful to corroborate. I am looking for a solution that clears all weights between iterations of a hyper-parameter search. I am running individual models as sub-process’.
chk_dir = '/root/.cache/torch/hub/checkpoints/'
if os.path.isdir(chk_dir):
for chkpnt in os.scandir(chk_dir):
print(f'rm"ing {chkpnt.path}')
os.system(f'rm {chkpnt.path}')