This is a collection of known issues which may arise when working with lambeq, including 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 NaN and 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]

When using 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 link, 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