Currently, when we give document ID as input to the collection module, a vector store gets screated and that vector store, along with the given model name and instruction creates an assistant and assistant id is what we get as an output here, with the name of the output variable being "llm_service_id".
Since Openai's assistant is going to get deprecated from their end, we should only be creating a vector store for the given documents in this collection module. Additionally, the module's code should be LLM agnostic and not be so specific to openai.
Additionally, these should be covered in the enhancement of the module as well -
-
DB session stays open during long external calls (e.g., creating OpenAI vector stores), holding connections/locks unnecessarily. more context : here and here
-
Need to take care of response payload, logs and read
-
Removing callback handlers specifically being used in the collection module, and just using the callback handlers in utils
Currently, when we give document ID as input to the collection module, a vector store gets screated and that vector store, along with the given model name and instruction creates an assistant and assistant id is what we get as an output here, with the name of the output variable being "llm_service_id".
Since Openai's assistant is going to get deprecated from their end, we should only be creating a vector store for the given documents in this collection module. Additionally, the module's code should be LLM agnostic and not be so specific to openai.
Additionally, these should be covered in the enhancement of the module as well -
DB session stays open during long external calls (e.g., creating OpenAI vector stores), holding connections/locks unnecessarily. more context : here and here
Need to take care of response payload, logs and read
Removing callback handlers specifically being used in the collection module, and just using the callback handlers in utils