Tal Almagor

TabNine will not be able to work with the jupyter extension [custom editor](https://github.com/microsoft/vscode-jupyter/blob/55ea387553812a3ca7c7b15bd1984a485fe37754/package.json#L1349). But good news - it does seem to work great with the native editor API which is...

encountered roughly the same, but the problem was "nan gradients" caused by layer norm, or more specifically something similar to - https://github.com/pytorch/pytorch/issues/41527. and with fairseq - https://github.com/pytorch/fairseq/blob/fcca32258c8e8bcc9f9890bf4714fa2f96b6b3e1/fairseq/modules/layer_norm.py#L38

Following @Facenomore23 comment i'd try - ```python from spacy_ke.util import registry, ngram_selection @registry.candidate_selection.register("custom") def ngram_selection(doc: Doc, n=3) -> Iterable[Candidate]: return ngram_selection(doc, n=2) # or any other logic ```