Now that we have somef extracting many fields from many files, I would like to explore using small, local LLMs (based on Ollama) for the following:
- Specific field metadata extraction. In particular, extended descriptions. Maybe installation instructions too.
- Author reconciliation (especially when the same information is in multiple sources)
- Requirement improvement (in particular when we have ranges or versions across multiple files)
Note:
- The LLM-based approach must be optional, as the framework is aimed to be lightweight.
- We need to have the basic functionality in
somef-core, have it released as a package and imported here.
Now that we have somef extracting many fields from many files, I would like to explore using small, local LLMs (based on Ollama) for the following:
Note:
somef-core, have it released as a package and imported here.