FANA LLM v0.1.92 Improved Database Retrieval and LLM Response Performance

We've just pushed a new release that bumps up the performance of our RAG database retrieval and LLM response times.

BACKEND

  • We have made some tweaks to the BE to squeeze out better performance from our database retrieval ops. Regarding performance, we're still rocking GPT-3.5-turbo-16k-0613 to orchestrate the backend workflow. However, if we notice the FE is consistently spinning up composite images or image-text combos, we'll likely need to migrate to GPT-4-turbo-preview - a major league performance hit to be sure.

Last updated