This is a collection of known issues which may arise when working with
possible workarounds. If you encounter a problem that is not listed here, we
encourage you to
submit an issue.
NaN and Inf errors during training
Since release 0.3.0,
lambeq provides its own numerically stable
loss functions which guard against
Inf numerical errors.
This change also removed safeguards from
QuantumModel, making protecting
against such numerical errors the responsibility of the loss function.
Any custom loss function must thus also guard against such errors.
SSL error [Windows]
lambeq <= 0.2.3 on a Windows machine, the instantiation of the
BobcatParser might trigger an SSL certificate error. If you require
lambeq <= 0.2.3, you can download the model through this
extract the archive, and provide the path to the BobcatParser:
from lambeq import BobcatParser
parser = BobcatParser('path/to/model_dir')
Note that using the
WebParser will most likely result in
the same error.
However, this was resolved in release 0.2.4. Please consider upgrading lambeq:
pip install --upgrade lambeq