NaNs in testing about pyxlstm HOT 3 CLOSED

the issue was resolved, I might do experiments and produce a document on the issue at a future point.

from pyxlstm.

Update

I have possible insights into what might be causing the issue, I have temporarily fixed the issue and got the training to work with the public dataset, however there is a small, a very small possibility this may trace back to the XLSTM implementation (I have to empirically test a few things before I post anymore updates.

from pyxlstm.

Thank you for the update on the issue. It's great to hear that you've found a temporary fix that allows training to work with the public dataset. I appreciate your proactive approach in investigating the root cause, especially the potential connection to the xLSTM implementation.
I look forward to your pull request with the temporary fix. When you're ready, please submit it with:

1. A clear description of the changes made
2. Any relevant test results or logs
3. A brief explanation of how it resolves the current issue

from pyxlstm.