A linter for deep learning non-determinism, randomness and reproducibility with applications to scientific computing and simulation.
From your terminal, clone this repository and move into the project directory as shown here:
git clone https://github.com/minnervva/torchdetscan.git
cd torchdetscan
Install Python from https://www.python.org or by using various other methods. After installing Python, create and activate a virtual environment within the project directory as follows:
python -m venv .venv
source .venv/bin/activate
Install the torchdet package and its dependencies into the virtual environment:
pip install .
Check installation by running the torchdet --help
command. This should output the following text in your terminal:
usage: torchdet [-h] [--verbose] {scan,test} ...
Find non-deterministic functions in your PyTorch code
options:
-h, --help show this help message and exit
--verbose, -v enable chatty output
subcommands:
{scan,test} valid subcommands
scan run the linter
test run the testing tool
Run the scan command line tool:
torchdet scan examples/pytorch_basic.py
Run the test command line tool:
torchdet test convolve2D
Use the --help
option with each subcommand to see more information about that command.