Introduced a way to mix and match embedding.
Only marginal improvement with Sentence Embedding.
todo:
clean up the config mess. new var EMBEDDING introduced
1. Move back to llm qa "stuff" mode as map_reduce is clueless
2. Created more metadata around documents.
* allow create different collection spaces for db.
/v1/admin/db -d "name={collection}"
* within collection allow file names so that answers can be annotated
{content: {answer :"I don't know", files:["a1","a2"]}}
would mean openAI doesn't know about the question even though
embedding returned two files.
Lot more work to do