The idea is:
Add either a section for code manipulation of embeddings in the embedding tool, or simply add a “convert embeddings to binary” toggle in the embedding tool.
My use case:
embeddings shrink in storage size, and look up speed is improved. All embedding use cases scale better
I think it would be beneficial to add this because:
scaling, speed, resource usage of database services like supabase (which supports this for stage and lookup).
Any resources to support this?
Are you willing to work on this?
willing and able are two different things. I’ll do what I can, but that won’t be much I suspect. The quantization code is already well documented in those articles and their linked examples, it’s just a matter of making a place for it to go between the embedding tool and whatever node it is connecting too. I suppose this could also be addressed by simply allowing the code tool to connect to the embedding tool port on nodes. Then that could be used to call the embedding endpoint and process the results before passing them back to the parent node.