also.. had to delete and recreate all my DB saved embeddings due to the upgrade, what changed? the previous embeddings worked great but of course would not work with the new 4.19 version, have we lost something in the refactoring/newer version?
The openai playground still works well for my use case on davinci, this seems more like the embedding search returned 'text chunk' not containing the correct information... i.e. I know the keyword is there in the embedding however it's not creating a useful answer as frequently. This def started happening around the time of the gpt version upgrade for turbo, but after the llama rename.
Is it my imagination or am I getting worse results with embeddings and queries EVEN with davinci. The previous versions of GPT-Index seemed magical, however the queries against generated embeddings seem to result in "The context information does not mention anything about...."... did the internal prompts change or should i use some optimized settings for the Vector creation, I've been using the library for over a month and def see a difference. hopefully a simple fix. I'm using documents = SimpleWebPageReader(html_to_text=True).load_data([url]) and index = GPTSimpleVectorIndex(..)
