NOTE: SETTING THE MODEL VIA GUI IS NOT YET SUPPORTED. This project is in active development. Please stand by or help by contributing!
Metadata Inference
The fallowing allows you to reconfigure, how metadata about your documents is inferred. This includes title, author, publisher.
This option infers the title, etc. based only on the PDF's metadata. It is important to check back to see the output before committing the changes to the library. While efficient, this method is approximate. There are likely mistakes.
Optionally, this process can be augmented with the Crossref database. This only works for scientific documents. The assumption is that the PDF is named after the ISBN for now. That way, it can look up the relevant information.
This uses a donut model finetuned on DocVQA..
It's peak memory usage is around 3 GB. However, it is fairly inaccurate and struggles with information that is written accross multiple lines. If possible, it is recommended to use the Idefics 2 model. See next option.
This is the best open source option right now. This option is in development. Please stand by.
The option is not yet supported.
Keywords inference
This is how the keywords are inferred. This option cannot be modified yet.
SmartFileLibrary - This project is in active devlopment.
Username