swalot 0.0.3

No CUDA out of memory again; fully use of a GPU card per user.

Installation

Install the package using pip:

pip install swalot

If you are using pip mirror, you may get error logs such as "No matching distribution found.." Try using pypi as the installation source:

pip install swalot -i https://pypi.python.org/simple

Usage

Simply import and wrap training code:

import swalot as sw

with sw.protect():
    """
    use CUDA tensor calculation here as usual.
    All RAM will be protected automatically!
    e.g.
    """
    # a = torch.randn(1000, 1000, 600).cuda()