Hi,
Background thread: How are text chunks for embeddings generated?
Can we have more granular control in the Retool Vector query? For example, when we choose to upload a document to a vector, it would be nice to have options to decide the splitters, maximum chunk size, etc. Essentially, it would be an explicit call of the chunkText and TextSplitter.
Highly needed!! would love to have something like this.
Hi there! Thanks for the request. I've passed this feedback on internally as well. Could you share more details on how you'd like to use this?
Are you looking to upload part of a document? Are you experiencing performance issues?
Hi @Tess,
Thanks for getting back. Essentially, I'd like to have options to set parameters such as chunkText and TextSplitter when I create a Retool Vectors Resource, either to upsert doc or insert doc, etc.
What I am doing now is once I have the OCR text from the document I want to upload, I would go to Resources -> Retool Vectors, then click "Create Vector". Since I can't customize the chunking strategy, I would need to manually edit the texts in a text editor, then upload the texts to create a vector through trial and error to get to the chunks I want. Ideally, when we are able to set the parameters, I don't need to go through the trial and error part any more.
Thank you for sharing these details! Will follow up here with any updates I get internally
On a related note, can we have the option to connect to other vector stores? I know there are a ton of them, so not necessarily looking for a one-click solution, but at least something that would allow the use of a different store and retrieval approach for RAG within the pre-built component.
Sorry for the delay, @jg80 ! Any specific integration? I checked internally & found out that we are looking at building integrations with other vector db's (but no eta yet)
I would want to be able to connect to a hosted Postgres DB extended with pgvector and Pinecone.