Try not to fail when there should be memory available#2869

awni merged 3 commits into

ml-explore:mainml-explore/mlx:mainfrom

awni:improve_cuda_allocatorCopy head branch name to clipboard