By default, all configs are set with extra="allow"``. To facilitate building the vocab with a full train config, e.g. inwiki_103orwmt17, we'll be more permissive forBuildVocabConfig`.
Note: pydantic 2.8.0 upgrade triggers errors that were not raised before. This outlined some remaining issues around some vocab/data validation logic. This PR provides some patches, but this logic might be reviewed more in depth at some point.
A somewhat cleaner way might be to nest the VocabConfig in a vocab key, and pop it when needed (like already done for data in some places).
By default, all configs are set with
extra="allow"``. To facilitate building the vocab with a full train config, e.g. in
wiki_103or
wmt17, we'll be more permissive for
BuildVocabConfig`.Note: pydantic 2.8.0 upgrade triggers errors that were not raised before. This outlined some remaining issues around some vocab/data validation logic. This PR provides some patches, but this logic might be reviewed more in depth at some point. A somewhat cleaner way might be to nest the VocabConfig in a
vocab
key, and pop it when needed (like already done fordata
in some places).