Known Issues#
Dataset Generation#
It is only possible for packages which are installed in current environment.
It expects the package to be thoroughly documented in
numpydocstyle.
Database Generation#
It does not capture parent-child information between source documents and smaller chunks that are embedded to match smaller chunk for relevance and use full document as context.
Information Retrieval#
It is sometimes non-deterministic.
It is often quite slow, especially when
MMRis used.
Response Generation#
It often gets stuck and takes long time to complete.
It sometimes fails completely with out of memory, but this is likely to vary based on device specifications.