Integer to NaN error with download_and_train.py

[mattpharris]

Description

Trying to successfully install+test the latest version of the repo.

Initially, I was hitting versioning errors if I just set the environment using:
conda create --name <my_env> python=3.6

So, I found the following commands allowed me to proceed without any errors:
conda create --name <my_env> python=3.6 python-dateutil=2.8.0 tifffile=0.15.1 scikit-image
git clone https://github.com/AllenCellModeling/pytorch_fnet.git
cd pytorch_fnet
pip install .
conda install quilt3 #found this wasn't installed with the previous line...
export MKL_THREADING_LAYER=GNU #my solution to the previous issue @Li-En-Good "Could not train the model on AWS #173"

After getting no errors from the above code,
cd examples
python download_and_train.py

The manifest downloads from quilt successfully, but I continue with the following NaN-to-integer error:

Any advice?

Expected Behavior

python download_and_train.py runs without errors.

Reproduction

Environment

Yes, I do have an NVIDIA graphics card

[alxndrkalinin]

See this solution